Galaxy evolution
Having seen what constitutes our Milky Way, i.e. ‘what is there’, we will now briefly explore the question how ‘it got there’, i.e. how a galaxy forms and evolves. Clearly, the field of galaxy formation and evolution is huge, and far beyond the scope of this chapter. Thus, we will only sketch the barest, qualitative outlines of the most crucial concepts and results.
The study of galaxy evolution has gained enormous momentum since very deep observations of almost randomly selected, small areas of the sky at a number of wavelengths have become available. The most famous of these ‘Deep Fields’ are the Hubble Deep Fields (HDF North, observed in 1995, and HDF South, observed in 1998, see Ferguson et al 2000 for a review of the HDF observations and their impact). Deep fields, often centered on the HDFs, have now been obtained in a multitude of wavelength ranges, i.e. in the far infrared (ISO, e.g. Serjeant et al 1997), in the submillimetre (Scuba, e.g. Hughes et al 1998) and at X-ray wavelengths (Chandra: Giaconni et al 2001).
In principle, these images allow the observation of galaxy evolution ‘as it happens’, since they contain both very distant and more nearby objects. Thus, it is possible to follow the evolution of the galaxy population. Of course, to do so, it is necessary to distinguish between true evolutionary effects and morphological variations. Thus, a measure of distance, or redshift, is needed. It is difficult or impossible to determine spectroscopic redshifts for very faint objects, but the problem has been moderated with the advent of much more easily measured photometric redshifts. It could be established that these redshifts are indeed fairly accurate (e.g. Fernandez-Soto 2001).
Hierarchical, bottom-up structure formation
The investigation of large-scale structure formation is the domain of numerical simulations of dark matter particles, using some of the largest parallel supercomputers in existence. For example, the ‘Hubble Volume’ project of the international Virgo collaboration (see www.mpa-garching.mpg.de/∼virgo/virgo) follows 10⁹ dark matter particles from z ∼ 5 (z ∼ 1.4 for a somewhat smaller volume) to the present. Thus, the development of galaxy clustering (Evrard et al 2001, Colberg et al 2000) or the mass function of dark matter halos (Jenkins et al 2000) can be investigated.
The simulations reproduce observational redshift surveys such as the 2dF survey (Percival et al 2001), the Las Campanas redshift survey (e.g. Shectman et al 1996) and soon the Sloan Digital Sky Survey to a reasonable degree of accuracy. At high redshifts, they show little large-scale structure, though of course the seeds of structure can be tied in with the Cosmic Microwave Background fluctuation spectrum, and beyond, as established by the COBE satellite and the recent Boomerang (e.g. de Bernardis et al 2000) and Maxima (Hanany et al 2000) balloon experiments.
The development of filamentary structures (‘bubbles and voids’) is striking and can already be followed in simulations with somewhat fewer particles, like the one by Jenkins et al (1998), based on 1.7 × 10⁶ particles, which was carried out assuming four different cosmological models (see figure 1.1). Universes with low matter content (Ω_M = 0.3) such as the currently favoured ΛCDM cosmology form structure earlier than universes with Ω_M = 1, like the formerly ‘standard’ Einstein–de Sitter universe, that seems to be ruled out by current data. At low redshift, all simulations give similar results. They probe size scales down to ∼10 kpc. Thus, they include the strong clustering regime and follow the hierarchical formation of clusters, but do not resolve the evolution of individual galaxies. The simulations have predictive power and help decide between cosmological models; for example an analysis of the Hubble volume leads to the prediction that too many hot (in the sense of X-ray temperatures) clusters like the Coma cluster at z ∼ 1 would not be compatible with a ΛCDM based model.
On the slightly smaller scale of the evolution of galaxy clusters, we expect the cluster environment itself to influence the evolution of the member galaxies. Smaller building blocks are expected to merge into larger objects, and this process can again be followed in simulations (e.g. Moore et al 1999). It turns out that, while hierarchical merging certainly occurs, the extent of the merging depends

Figure 1.1. Four simulations of large-scale structure formation by the VIRGO consortium (Jenkin et al 1998). Four different cosmological scenarios are assumed; the development of structure is depicted at redshifts z = 3, 1, and 0. The box size is 239.5 Mpc/h.
on the velocity profile and dispersion of the cluster. In any case, small dark matter subhalos seem to survive in surprisingly large numbers down to the scale of individual giant galaxies. In simulations, the halos of individual galaxies look very similar to galaxy clusters, and this result persists at the highest currently possible resolution of numerical, dissipationless N-body simulations (Ghigna et al 2000).
Thus, hierarchical models for structure formation naturally form massive dark matter halos with a wealth of substructure. On a cluster scale, this substructure can easily be identified with visible galaxies, that are (with some bias function that is difficult to determine) hosted by the dark matter subhalos. The observed distribution of galaxies in clusters agrees reasonably well with prediction.
On the scale of individual galaxies, however, there may be a problem: here, the subhalos correspond to satellite dwarf galaxies, and the observed abundance of such dwarf satellites does not match the large number of predicted dark matter subhalos (Moore et al 1999, Klypin et al 1999). It remains to be seen whether this is a serious problem of the otherwise very successful Cold Dark Matter (CDM) simulations or whether it can be solved by fine-tuning the models, as suggested by, e.g., Font et al (2001), who argue that at least the dynamical impact of the subhalos on the thin disk of a galaxy should be minor.
Evidence for hierarchical, bottom-up galaxy formation originates not only from numerical simulations; it is strongly supported from an observational viewpoint. Analysis of deep field data shows that the morphological classification of galaxies by the classical Hubble sequence breaks down at redshifts >1 (e.g. Driver et al 1998). Barred galaxies seem to become rare at even lower redshift (Abraham et al 1999) while the number of ‘irregular’ or ‘peculiar’ galaxies increases steeply (see the recent review by Abraham and van den Bergh 2001).
In general, the size–redshift relation seen for E/S0 and spiral systems seems to point to their assembly of redshifts >1. There also seems to be an excess of faint blue, very compact galaxies, many of which are located at z ∼ 0.5, but some are at z > 2. This appears to be an actively evolving galaxy population, constituting, at least in part, the building blocks of larger systems.
In some cases, these subsystems or building blocks may have been caught in the act of taking part in a hierarchical merging process. Pascarelle et al (1996) found 18 small, bluish objects in a ∼0.7 Mpc field at z = 2.39 and these were interpreted as the building blocks of a future large galaxy. Similarly, Campos et al (1999) reported the detection of 56 Lyman α emitters in a small field adjacent to a quasi-stellar object.
Evolutionary mechanisms: mergers and ‘internal’ processes
Most galaxies are not isolated as we have seen, they tend to form in clusters, or at least in groups. This cluster environment is expected to influence not only the number of galaxies directly through merging, but also their type. This effect, dubbed ‘galaxy harassment’, probably has consequences for the balance of Hubble types in a cluster between redshifts of, e.g., z ∼ 0.4 and the present, that is, in a redshift regime where the Hubble sequence still describes the galaxy type adequately. It seems that more distant clusters have a larger relative fraction of small spiral galaxies, many of which show some indication of high star formation or starburst activity, than clusters in the local universe. The latter are dominated by spheroidal galaxy types, ellipticals and S0s (Moore et al 1998).
Merging itself is, of course, the most dramatic and obvious driving force of galaxy evolution. In a scenario of hierarchical structure formation, the merger rate is expected to increase with redshift proportional to (1 + z)m with m = 2–3. This relation is expected to hold at least up to z = 2–3.
There are many excellent simulations of galaxy mergers (e.g. Barnes and Hernquist 1996, Mihos and Hernquist 1996, Springel et al 2001). It is important for the outcome to take into account the role of gas and (if possible) the feedback of the star formation process (see Kauffmann et al (1999) for a prescription). The simulation of collisionless dark matter or stellar particles can only be a first step in such simulations. The morphology of interacting and merging galaxies is well reproduced by models, down to the tidal tail structure of individual real-life objects. Usually, the models suggest that the merger remnant looks much like an elliptical galaxy. Specifically, the remnant structure follows the well-known R^1/4 law for the surface brightness of an elliptical fairly well, though it somewhat depends on the initial conditions and there may be discrepancies in the details.
This has led to the conclusion that ellipticals are the endpoints of galaxy evolution through mergers. During the merger, the galaxies pass through a phase of a very intense central starburst, since the gas is concentrated quickly into the nuclear region. Briefly, they may shine brightly as ULIRGs (ultra-luminous infrared galaxies), emitting more than 10¹²L_ּ in the Far Infrared. During this phase, most of the gas of the progenitor galaxies is consumed. Observations have indeed shown that all ULIRGs seem to be mergers, often even multiple ones. Once the starburst is over, ellipticals (and spiral bulges) evolve only passively, i.e. by the ageing of the stellar population. If they have largely assembled in a ‘merger age’ at z ∼ 2, this explains naturally why most ellipticals and bulges today appear to be old, reddish objects.
Apart from the spectacular evolution by merging, ‘internal’ mechanisms may also lead to secular changes in the appearance of galaxies. These are slower evolutionary mechanisms that take place after the initial assembly and most easily work on galaxies which have not become ellipticals, but are instead gas-rich disk galaxies. Interactions which do not lead to mergers play an important role in triggering or at least speeding up these evolutionary processes.
In contrast to the passively evolving ellipticals, the disk galaxies continue to form stars at a fairly constant rate of a few M_ּyr⁻¹. This star formation takes place, as we have seen, mostly in spiral arms, and possibly in the central region, especially if it is fed by a bar. The detailed structure of the spiral arms themselves is almost certainly subject to secular changes, even in Grand Design spirals, though the general character of a disk galaxy as a spiral remains unchanged. What is the gas supply for continuing star formation? Some of the necessary replenishment takes place by inflow from the more gas-rich outer regions of the disk we have seen that the H I disk often extends far beyond the optical disk. It is, however, also possible that infall of intergalactic (intra-cluster or intra-group) or halo gas clouds takes place. Chemical evolution models and the star formation and thus the gas consumption rates in many disk galaxies may require some infall.
Evolution along the Hubble sequence may happen to some degree. If so, then the direction of evolution is late → early, since all processes result in a higher

Figure 1.1. Left panel: the space density of a quasar has a maximum at a redshift of   ∼ 2 (from Shaver et al 1999). Right panel: the same may be true for the star formation rate, but in this case, the decline beyond z ∼ 2 is not well established (Cattaneo 2001; the points with error bars correspond to measurements; the curves are predictions for different evolutionary models).
central mass concentration and thus a more pronounced bulge. A bar, whether transient or persistent, should usually be involved in such a process, since it is the most efficient means of angular momentum transport.
More dramatic events, i.e. major mergers between disk galaxies with a resulting strong starburst, may of course also take place in the local universe, and we know a number of examples (e.g. Arp 220 and similar objects). Major mergers are, however, rare in the present-day cosmos. Accretion of smaller galaxies is a far more frequent process. In such a ‘minor’ merger, a large disk (or elliptical) galaxy swallows a smaller companion. It is likely that the Milky Way has been involved in several such acts of cannibalism during its history. At present, it is performing another one: it is in the process of consuming a small dwarf galaxy, the Sgr dwarf, which has already been disrupted and stretched out to a degree that made its very detection difficult (Ibata et al 1994).
The growth and evolution of a spiral bulge, e.g. through infall or inflow, may also have consequences for the central black hole, possibly by regulating its rate of mass accretion.
We have seen that the merging rate was certainly higher in the past, and may have had a maximum around z ∼ 2. Possibly related to this, there is undisputed evidence for a ‘Quasar Epoch’ at the same redshift (Shaver et al 1996, 1999). Quasars are thought to be powered by the most luminous supermassive black holes in the universe. Not only was the true quasar space density at z ∼ 2 more than an order of magnitude higher than it is now, it is also clear from, e.g., complete samples of radio-loud quasars that it declined rapidly at redshifts > 2.5 (see figure 1.1, left-hand panel).
Investigations of the star formation history of the universe, pioneered by Madau et al (1996), also show a clear rise by more than a factor of 10 from the present to z ∼ 1. It is less clear, however, whether the star formation rate declines at z > 2–3 (e.g. Cattaneo 2001, see figure 1.1, right-hand panel). This depends, among other factors, on the role of dust extinction in the high-z starburst galaxies and how a population of extremely dusty starbursts that appears in sub mm wave deep fields, and seems to have enormous star formation rates, is taken into account.
In any case, the rise in the quasar space density, the merger rate and the star formation rate almost ‘in lockstep’ is very suggestive of a scenario where the quasar brightness is explained by high accretion rates on massive central black holes, which were assembled at roughly the same time in frequent mergers of galaxy bulges, going along with massive starbursts. Consequently, the black holes experienced the bulk of their growth in this period, and might have grown only slightly since their initial assembly. This is indicated by the strong decrease in quasar light originating in the local universe.