Ten Billion Years of Galaxy Evolution

As paleontologists dig down to deeper layers in the Earth, they encounter ever older fossils. By the same token, astronomers can look back in time by observing galaxies at ever larger redshifts. The high resolution provided by the Hubble Space Telescope (HST) makes it possible to study the evolution of the structure within galaxies over timescales approaching the age of the universe. Astronomers can now explore the evolution of morphological classes of galaxies in ways that are similar to those in which paleontologists study the evolution of fossil species and genera over time. Perhaps the most unexpected result of such work is that the Hubble tuning‐fork diagram provides a satisfactory framework only for the classification of galaxies with z≤0.5, i.e., for galaxies with ages¹ less than half the age of the universe. Furthermore, with increasing look‐back time, it becomes ever more difficult to shoehorn galaxies into the morphological sequence Sa–Sb–Sc.

It is the purpose of the present paper to examine the details of how the Hubble classification scheme breaks down with increasing redshift. Part of the evolution of galaxies over the last 10 Gyr is, no doubt, due to internal factors, whereas the rapidly growing frequency of mergers with increasing redshift is beautifully visible on HST images. Such growth in the merger frequency with increasing redshift, which is expected from hierarchical merger models (Carlberg 2000), provides an external contribution to the secular changes of galaxy morphology. Strong observational support for a rapid increase in the merger and accretion rate of galaxies toward larger redshifts is provided by the present data and by observations of Patton et al. (2002).

The Ohio State Bright Spiral Galaxy Survey of Eskridge et al. (2002) provides a more or less unbiased (van den Bergh et al. 2002) sample of nearby spiral galaxies. After degrading 101 of these images to the resolution and signal‐to‐noise ratio (S/N) of galaxies at z = 0.7 in the Hubble Deep Field (HDF), van den Bergh et al. (2002) find that only 12% of these objects at z ∼ 0.0 are classified as being peculiar. Such peculiar galaxies are, by definition, galaxies that do not fit comfortably into the Hubble classification scheme. This observed fraction of peculiar objects among nearby spirals is significantly lower than that encountered in the HDF and its Flanking Fields (FFs), in which van den Bergh et al. (2000) found that 17 out of 37 (46%) of spirals with 0.60<z<0.89 are peculiar. In other words, almost half of all spiral galaxies appear to be peculiar at look‐back times of ∼8 Gyr. This comparison is almost independent of bandshift effects because the ground‐based z = 0 images were classified in B light, whereas the classifications of galaxies at z ∼ 0.7 were made in the infrared F814W band. In other words, the galaxies at z = 0.0 and z = 0.7 were classified at almost the same rest wavelength.

Van den Bergh et al. (2000) show that distant early‐type galaxies are less likely to be peculiar than is the case for galaxies of later morphological types. Only 1/22 (5%) of E–Sa–Sab galaxies were observed to be peculiar at z ∼ 0.7, compared to 11/16 (69%) of Sb–Sbc galaxies. These figures probably understate the difference in the fraction of peculiar early‐type and late‐type galaxies. This is so because a significant fraction of "Sc" galaxies are so peculiar at z ∼ 0.7 that many of them have been classified as "proto‐Sc," "Pec," or "?". These results suggest that compact early‐type galaxies started to approach their "normal" present‐day morphology faster than did more extended late‐type spirals. However, a caveat is that it might be more difficult to see peculiarities in distant compact early‐type galaxies than in equally distant (but more open) late‐type spirals.

Hubble (1936) noted that nearby spiral galaxies fall naturally into two sequences, which he dubbed "normal spirals" (Sa–Sb–Sc) and "barred spirals" (SBa–SBb–SBc). One of the first surprises (van den Bergh et al. 1996) provided by inspection of the HST images of the Hubble Deep Field–North (HDF‐N; Williams et al. 1996) was that the frequency of barred spirals appeared to decrease with increasing redshift. This conclusion was subsequently confirmed in the HDF‐N and HDF‐S by Abraham et al. (1999). However, before accepting this conclusion, one has to test whether the apparent decrease in the frequency of bars with increasing redshift might be due to (1) diminishing resolution with increasing redshift and/or (2) a decrease in the S/N with increasing z. Furthermore, such investigations have to be carried out at constant rest wavelength to guard against a possible wavelength dependence of the frequency of bars (Eskridge et al. 2000). Such an investigation has recently been undertaken by van den Bergh et al. (2002), who compared the fraction of barred objects among a representative sample of nearby spirals observed in blue light with the fraction of galaxies classified as SB in the same sample of galaxies, but with resolution and S/N degraded to that for HST images at z ∼ 0.7. To compensate for bandshift effects, these degraded images were classified in the HST F814W band. The data showed that ∼2 / 3 of the strongly barred (DDO types SB0–SBa–SBb–SBc) galaxies could still be recognized as barred spirals at z ∼ 0.7. In other words, most of the observed decrease in the frequency of SB galaxies between z = 0.0 and z = 0.7 is intrinsic and not due to the effects of a decrease in resolution and noise with increasing redshift. However, it should be emphasized that such resolution and noise problems might render some bars (if they exist) unobservable at z≫0.7.

Table 1 lists the fraction of all spirals (including objects of type S0) with known redshifts that are barred in the HDF‐N from van den Bergh et al. (2000) and in the HDF‐S from S. van den Bergh & R. G. Abraham (2002, in preparation). These data show that the fraction of barred [S(B) + S(B?) + SB] spirals decreases from ∼23% for galaxies with z<0.5 to ∼4% for galaxies with z>0.7. As has already been noted above, all of these classifications were made at, or close to, the ground‐based B band. Taken at face value, these results suggest that the frequency of barred spirals at z>0.5 might be lower than it is in nearby regions of the universe. If confirmed by future observations, this result could be explained by assuming that the disks of spirals at large look‐back times are still too chaotic (dynamically hot) to be susceptible to global barlike instabilities. Such global dynamical instability of disks is currently (Sellwood & Willkinson 1993) the favored formation mode for bars in spiral galaxies. The notion that galaxies at large redshifts are more chaotic than nearby ones is also supported by the work of Reshetnikov et al. (2002), who find that disks at large redshifts are (on average) more often warped than are the disks of nearby galaxies.

It is noted in passing that only ∼40% of strongly barred galaxies with I(F814W) < 22 mag, degraded to the resolution and S/N of the HDF FF (corresponding to an integration time of two orbits), were recognizable as barred spirals. This clearly shows that the low S/N in the Flanking Fields makes it more difficult to recognize barlike features in distant galaxies.

In the original version of this paper, it was noted that the HDF‐S galaxy J223302.81−603325.2, at a photometric redshift z = 1.578, is the only presently known barred spiral at a large redshift. However, a more recent photometric redshift of this object that is posted on the Stony Brook Web site¹ appears to show that the photometric redshift of this object is actually only z = 0.76. At this much lower redshift, it is less surprising to find a barred spiral.

Luminous "grand‐design" spiral galaxies of DDO luminosity classes I, I‐II, and II have large dimensions and are therefore relatively easy to recognize at great distances. The combined data for the HDF‐N by van den Bergh et al. (2000) and for the HDF‐S by S. van den Bergh & R. G. Abraham (2002, in preparation) contain classifications for 110 spirals of known redshift, of which only seven are grand‐design objects. None of these grand‐design spirals has z>0.7. Possibly this lack of very distant grand‐design spirals is linked to the observation that the spiral structure of late‐type galaxies appears to become increasingly chaotic as one moves to larger redshifts. However, it is not yet safe to conclude that grand‐design spirals exist only at low redshifts. This is so because a Kolmogorov‐Smirnov test on the small samples described above shows no statistically significant difference between the redshift distributions of ordinary spirals and of grand‐design spirals. So it is not yet possible to say at which epoch luminous disk galaxies started to form their grand‐design spiral features. It would clearly be of great interest to look into this question in more detail by obtaining redshifts (and accurate morphological classifications) for much larger samples of distant grand‐design spirals. A possible complication is provided by the fact that some interacting objects at z>0.7 exhibit tidal features that might be confused with the chaotic spiral arms that are diagnostic of the most distant known spiral galaxies.

A number of authors (Brinchmann et al. 1998; Lilly et al. 1995, 1998; van den Bergh 2001) have considered the possible evolution of the frequency distributions of galaxy types over time. In particular, Lilly et al. (1995) and van den Bergh (2001) concluded, from observations of galaxies in the Canada‐France Redshift Survey, that there is very little change in the density (or in the luminosity) of the reddest (early‐type) galaxies over the entire range 0<z<1. However, this conclusion should be treated with caution because changes in the space density (and hence galaxy morphology along any particular line of sight) may be affected by the well‐known correlation (Dressler 1980) between the frequency of classification types and space density. To avoid such problems in the future, it will be necessary to undertake deep homogeneous redshift and classification surveys of numerous galaxies at many positions on the sky. Only in this way will it eventually be possible to gain better insight into systematic changes in the relative frequency distributions of different galaxy types with look‐back time.

Both the HDF‐N and the HDF‐S are crowded fields. It is therefore expected (and spectroscopy confirms) that a significant fraction of all close binary galaxies are optical, rather than physical, pairs. Galaxies were therefore classified as a "merger" only if (1) their main bodies overlapped or (2) their outer structures appeared to show evidence for physical overlap or tidal distortion. Since faint outer tidal features are probably "lost in the noise" in the short‐exposure Flanking Fields, only classifications in the Hubble Deep Fields themselves (van den Bergh, Cohen, & Crabbe 2001; S. van den Bergh & R. G. Abraham 2002, in preparation) were used to estimate the fractions of merging galaxies that are listed in Table 2. These data show that about 5% of galaxies with z<1.2 in the Hubble Deep Fields appear to be mergers. For comparison (see Table 2), S. van den Bergh & R. G. Abraham (2002, in preparation) find that 62 out of 109 (57%) of their sample of HST galaxies with (mostly photometric) redshifts of z>2.0 appear to be mergers. In other words, the typical galaxy at z<1 (age younger than 8 Gyr) is single, whereas the average object at z>2 (age older than 10 Gyr) appears to be merging.

The majority of galaxies with z<1 are observed to be disk galaxies. On the other hand, a large fraction of the galaxies with z>2 are best described as resembling centrally concentrated "blobs," many of which appear to be interacting (or merging) with other blobs. Probably these blobs are the progenitors of our present‐day galactic bulges and elliptical galaxies. This speculation is supported by the observation that the objects seen at z>2 typically have quite small linear dimensions (Ferguson, Dickinson, & Williams 2000). Furthermore, fitting of multicolor observations to evolutionary models (Popovich, Dickinson, & Ferguson 2001) suggests masses of 10⁹–10¹¹ M_⊙ for the luminous inner regions of Lyman break galaxies (LBGs). Such values are diagnostic for the cores of galaxies rather than rather than of giant young star clusters. Clustering properties of LBGs have so far placed only weak constraints (Primack, Wechsler, & Somerville 2002) on the masses of the halos of Lyman break galaxies.

7.1. Morphology at z>1

7.2. Morphology at z>2

The recent NICMOS H₁₆₀ images of galaxies with 2<z<3 by Dickinson (2000) show that (after correcting for a difference in linear resolution) the morphology of galaxies is surprisingly similar at λ₀ = 1700 and at λ₀ = 4300 Å. This result suggests that the apparent structure of the images of such a distant galaxy is mainly determined by their intrinsic morphology and is not a strong function of bandshift effects. This result shows that (1) the stars that dominate the light at λλ₀ = 1200–1800 Å also dominates at λλ₀ = 4000–5500 Å; (2) if components of differing age are present, then they must usually have a rather well mixed spatial distribution; and (3) dust extinction cannot be the dominant factor determining the overall structure of most galactic images at high redshift. In view of these results, it appears safe to neglect bandshift effects for first‐order studies of the morphology of galaxies with z>2.

Many of the galaxies at z>2 are so different from those at low redshifts that it would be quite meaningless to assign Hubble types to them. Proceeding in an entirely pragmatic fashion, S. van den Bergh & R. G. Abraham (2002, in preparation) have established an empirical classification scheme for these objects, which is illustrated in Figure 1. The tentative classification system of van den Bergh & Abraham has two parameters. The first of these indicates whether a system appears to be single (s), binary (b), or multiple (m). It should, of course, be emphasized that such a classification system is somewhat arbitrary because a member of a wide binary might reasonably also be considered a single. Furthermore, a system with two bright and one faint components might be classified as a binary, rather than as a multiple system. The second classification parameter describes the structure of the brightest component of the system. Small, compact, circular knots are denoted by "1," slightly fuzzy brightest components are classified "2," fuzzy images with comma‐like structures are assigned to class "3," and tadpole‐like objects with longer tails are designated type "4." Finally, chain galaxies (Cowie, Hu, & Songaila 1995) are assigned to class "5." It is not yet clear if the formation of such galaxy chains is in any way related to the filamentary structures that are found in numerical simulations of the early evolution of cold dark matter. It, perhaps, should be emphasized that the classification types "1," "2," and "3" refer to the structure of the brightest knots within structures, which may themselves be quite chaotic.

Zoom In Zoom Out Reset image size

In the nomenclature adopted in the present paper, "s1" denotes a compact single object, "b2" denotes a binary with a slightly fuzzy brightest component, and "m3" denotes a multiple object with a comma‐like brightest component. For objects of types 4 and 5, the s, b, and m designations are not appropriate. It is not yet clear if the strictly empirical system described above correlates with luminosity, integrated spectral type, or any other physical characteristic of the objects being classified. In this connection it is, however, of interest to note that all nine of the most distant (z>3.5) objects classified by van den Bergh & Abraham are single objects of types s, s1, or s2. The lack of outer "fuzz" in these objects might, at least in part, be due to the effects of cosmological (1 + z)⁴ dimming of the faint outer regions of these "galaxies." From a morphological point of view, the most distant objects in the HDF‐N and the HDF‐S might also be assigned to Hubble types E0–E3 or Sa. It is of interest to note that none of the objects with z>3.5 appear to be as flattened as edge‐on spirals or E4–E7 galaxies. This result is somewhat puzzling because widely accepted cold dark matter evolutionary scenarios (e.g., Kauffmann, White, & Guiderdoni 1993; Baugh et al. 1998) have ellipticals forming from merging disk galaxies. Why do we not observe such disks at large look‐back times? Possibly (Weil, Eke, & Efstathiou 1998) such stellar disk would be destroyed by frequent mergers at z≫2 and by stellar feedback processes that suppress cooling of gas before z ∼ 1. The destruction of disks by tides and mergers at large look‐back times might also account for the unexpectedly small half‐light radii of galaxies at z>1.5 (e.g., Roche et al. 1998).

If the speculation that the formation of disks is impeded by frequent mergers and tidal effects is correct, then one would not expect large spirals (such as the Milky Way) to contain a very old thin‐disk population component.

The main results of investigations of the dependence of galaxy morphology on redshift may be summarized as follows:

• 1.  
  A major transition in the dominant mode of galaxy and star formation is observed to occur between z ∼ 1 and z ∼ 2. At low redshifts, most star formation is presently taking place quiescently in well‐ordered disks. On the other hand, much of the star formation at high redshifts appears to be associated with chaotic mergers.
• 2.  
  This change might contribute to the change in the slope of the Madau plot that is observed at z ∼ 1.5.
• 3.  
  The majority of objects with z<1 are disklike, whereas most of those with z>2 appear to either be chaotic or have dominant bulgelike components. A nonexhaustive list of possible scenarios includes (1) the first disks were destroyed by frequent mergers, (2) the gas in these disks did not cool before z ∼ 1, or (3) early gaseous disks were destroyed by the strong winds resulting from massive starbursts.
• 4.  
  The Hubble tuning‐fork diagram provides a good framework only for the classification of galaxies with z≤0.5.
• 5.  
  Study of artificially degraded images shows that the fraction of all galaxies that are barred spirals decreases with increasing redshift. This effect is tentatively interpreted as being due to the fact that hot (chaotic) disks might not be able to develop global barlike instabilities.
• 6.  
  The fraction of peculiar galaxies, i.e., objects that do not fit naturally into the Hubble classification scheme, increases from 12% at z = 0.0 to 46% at z ∼ 0.7.
• 7.  
  Compact early‐type (E–Sab) galaxies appear to approach their final morphology more rapidly than do more extended late‐type (Sbc–Sc) galaxies. As a result, only ∼5% of early‐type galaxies at z ∼ 0.7 appear peculiar, compared to 69% peculiars among late‐type galaxies.
• 8.  
  At look‐back times greater than 5 Gyr (z>0.5), the spiral‐arm structure of early‐type galaxies appears to be underdeveloped.
• 9.  
  At look‐back times greater than 7 Gyr (z>0.8), the structure of the spiral arms of late‐type galaxies appears less regular (more chaotic) than it does at lower redshifts. It should be emphasized that this is an easily observed gross effect that is not likely to be affected by either noise or image resolution.
• 10.  
  A tentative, and strictly empirical, classification scheme is proposed for those objects that are viewed at look‐back times greater than 10 Gyr (z>2).