A Holy Grail of modern astronomy is understanding the origin of Edwin Hubble's morphological sequence of galaxy types. What made some collapsing gas clouds turn into elegant spiral systems like our own Milky Way, whereas others became smooth, featureless ellipticals? More fundamentally, does the taxonomic scheme introduced by Hubble in the 1920s have any physical relevance? The Space Telescope that bears Hubble's name is providing answers to these and other questions in the context of modern theories of structure formation.
Edwin Hubble was a perceptive observer who introduced the morphological classification scheme for galaxies still in use today. His “tuning fork” diagram (figure 1) differentiates ellipticals from unbarred and barred spirals, further classified according to the tightness and fine structure of their spiral arms which appears to correlate with the fraction of light present in the central bulge. Intermediate to the spirals and ellipticals, Hubble also introduced the enigmatic lenticular or S0 class. These share the disk-like structure of spirals with an inner nuclear bulge, but are devoid of spiral arms and share the smoothness and colours of the elliptical population. Over 90% of nearby luminous galaxies can be categorized within Hubble's scheme, described in more detail by van den Bergh (1998).
Whilst it is natural to attempt to classify galaxies according to their appearance, does such a scheme offer any insight into the physical origin of the different types? Hubble evidently thought so: he referred to the Sc through Sdm types as “late type galaxies” and the spheroidals (a collective term for the ellipticals and S0s) as “early types”, imagining an evolutionary sequence along the tuning fork. Indeed, the integrated colours of galaxies of different types show a strong trend that can be explained in terms of their current star formation rate. Blue late-type galaxies can be understood as ones that are particularly active in forming stars, whereas the redder early types are more quiescent. However, this inference need not necessarily imply that one type evolves into the other. Rather the sequence might represent the extent to which the star formation declines from its initial value. Beatrice Tinsley, a pioneer of galaxy evolution models in the 1970s, analysed the optical-infrared colours of a sample of local galaxies and argued that continuously star-forming spirals would look much the same in the past, whereas the presently red ellipticals would, closer to their moment of birth, have been intensely luminous and blue, possibly detectable at large redshifts as young primaeval galaxies. By peering into the past, astronomers might be able to directly test Tinsley's evolutionary predictions and, most importantly, find examples of primaeval galaxies.
Structure formation and dark matter
Since the 1930s there had been suspicions that a large fraction of the matter in the universe is not in stellar or gaseous form. By the late 1970s and early 1980s, evidence for dark matter became overwhelming. Estimates of the gravitating mass in various regions could be contrasted with that responsible for the luminous component. Together with constraints on the amount of dark matter that could be made of baryons (protons, neutrons and related particles) from the abundances of light elements formed in the Big Bang, it became clear that much of the dark matter must be of a different, non-baryonic form. (The dark matter picture is described in more detail than below by Baugh et al. 1996, Cole et al. 1994, Frenk et al. 1985 and White and Rees 1978.)
This deduction transformed our view of how galaxies form. In order to explain the observed large-scale structure and abundance of galaxies, the cold dark-matter model posits that each visible galaxy is embedded in a much larger dark-matter halo of non-baryonic material, which detached from the expanding cosmic plasma at early times. The gravitational merging of these halos drives the growth of structure and the rate of cooling of hydrogen gas drawn into the merging halos governs the history of assembly and star formation in normal galaxies.
The dark-matter picture can successfully reproduce both the degree of fluctuations observed in the microwave background and correlations found in the local galaxy distribution. It predicts that dark-matter halos assemble hierarchically from smaller examples according to their mutual gravitational attraction. However, in describing how galaxies assemble, we are primarily concerned with the baryonic component and the rate at which pristine hydrogen gas converts to stars as well as, ultimately, with dynamical and structural details which would enable us to understand and differentiate spirals from ellipticals and bulges from spiral arms which themselves appear in barred and unbarred form.
Although promising progress has been made in simulating how galaxies might assemble in the dark-matter picture, via calculations of how gas cools and forms stars in assembling dark-matter halos, we are a long way from a detailed prediction of how Hubble's sequence emerged. For this reason we proceed along phenomenological routes. One of the most important issues we might hope to understand is whether the morphologies we observe today represent those established at birth, or whether transformations can occur. In the former case, galaxy morphologies would bear important clues to physical conditions at the epoch of formation (as well, perhaps, as pinpointing that epoch) whereas in the latter case, the abundances of the various types would constrain the history of the transformation processes.
Before the development of the dark-matter picture, most cosmologists believed that at least some galaxies (e.g. the presently quiescent ellipticals) formed via a rapid “monolithic” collapse at high redshift. Gas turning promptly into stars at the time of collapse would explain many of their observed properties. The high central stellar densities in local ellipticals would reflect the mass density of the universe at those early times, particularly if the protogalactic cloud collapsed by only a small amount. A modest collapse factor would also be consistent with the small angular rotation observed; the spin-up during collapse would be slight. By contrast, spiral galaxies might represent systems that collapsed by significant amounts, with star formation delayed until the gas clouds had collided many times and settled into a disk. Small amounts of initial angular momentum in the protogalaxy would amplify rotation during collapse, leading to rapidly rotating stellar systems (figure 1). The key factors differentiating the two classes are the collapse factor and the degree to which energy released during collapse is dissipated through gas cloud collisions. For ellipticals, there is no dissipation. Stars form in situ at the time of collapse, producing galaxies with smooth shapes. For spirals, significant dissipation implies large collapse factors, rapidly-rotating products and disk-like stellar distributions.
A central question is whether galaxies form and evolve in isolation. Unlike stars in our own galaxy, individual galaxies have large cross-sections compared to their typical separations and, consequently, collisions are expected to occur. At early times, the number of galaxies per unit volume was higher and so we can expect mergers to have been more frequent. Numerical simulations suggest dramatic changes can occur in such circumstances. Take a collision of two spiral galaxies as an example. In such a case, the delicate stellar disks are destroyed, the circular stellar orbits are randomized and the hydrogen gas is either expelled or converted into stars, leaving a smooth elliptical remnant (figure 1). Close examination of some (but not all) ellipticals reveals tell-tale remnants of recent merging. In order to prevent wholesale destruction of spirals by merging, dark-matter enthusiasts have suggested that large galaxies must have assembled quite recently so that many ellipticals are, in fact, the merger product of two spirals.
The rate of merging expected in the dark-matter models is sensitive to the details of the cosmological model. If, for example, we admit the presence of a non-zero cosmological constant as suggested by recent supernovae data, then the rate of merging expected within the redshift range probed by HST is modest. But if we live in a high density universe with no cosmological constant, a great deal of recent merging is expected.
In summary, the key difference between the traditional and dark-matter pictures for the origin of galaxy morphologies lies in the extent to which galaxies may have interacted during their history, as well as the extent to which it may be possible to have early “monolithic” collapse of large systems as opposed to more gradual, hierarchical assembly.
Studies of galaxy evolution using HST
Following the insertion of correcting optics in the instrumentation on board Hubble Space Telescope (HST), it became clear to our group at the Institute of Astronomy at Cambridge that we could directly witness the history of galaxy morphology by exploiting deep HST images spanning a substantial range in look-back time. Prior to HST, attempts had been made from the best ground-based observatories to resolve and classify distant galaxies but, even in the best conditions, galaxies as close as redshifts of 0.3–0.5 (corresponding to “look-back times” of 3–5 billion years in a universe 12–15 billion years old) were mostly blurs of light (figure 2).
In contrast, the best HST images permit reliable classifications to be made to epochs corresponding to redshifts of 1–2, some 60–70% of the way back to the Big Bang. Early work with HST concentrated on randomly chosen small fields observed with the Wide Field Planetary Camera 2 (WFPC-2) in parallel, while others used primary instruments to observe specific targets. A collection of about 100 such deep images comprised the Medium Deep Survey (), one of the early Key Projects of the telescope. The disadvantage of this dataset was the absence of comprehensive redshift data for the fields in question. The field of view of WFPC-2 is poorly matched to ground-based multi-object spectrographs and thus a second project, undertaken jointly with colleagues in France, Canada and Australia, consisted of producing a mosaic of WFPC-2 images of redshift survey fields. A substantial boost was later provided by the colour exposures taken as part of the two Hubble Deep Field campaigns (figure 2).
Comparing morphological distributions observed at different eras or faintness limits is fraught with possible pitfalls. Faint, distant galaxies are redshifted so that images taken in a given photometric band refer to ones viewed at shorter rest wavelengths where the light distribution in galaxies is governed by younger stars which clump in associations. So a galaxy may appear more patchy or irregular at faint magnitudes merely as a result of this effect. Likewise, lower surface brightness features such as spiral disks will rapidly be dimmed beyond detection at great distances and so certain classes of galaxies may appear more concentrated and dominated by their nuclear regions than would otherwise be the case. At first it was thought these effects would be insuperable, but in many fields HST has observed the same galaxies in different wavelengths so that the effects can be calibrated. Moreover, an industry of detailed numerical simulations has been developed, whereby multicolour images of local galaxies are reconstructed so as to mimic their appearance with HST at high redshift. To the accuracy required to compare the morphological distributions with predictions, the corrections for these effects can be adequately controlled (figure 2).
What then are the key questions that HST data should address in attempting to piece together a coherent picture for the origin of the Hubble sequence of galaxy types?
The faint blue galaxy problem
A long-standing puzzle in extragalactic astronomy has been the origin of the excess blue, star-forming galaxies seen in deep images. Since the late 1970s it has been known that there are many more such systems than expected from our knowledge of the locally observed population. One of the first results from the Cambridge HST surveys was the identification of a strong increase with redshift in the fraction of morphologically peculiar galaxies. The integrated luminosity density of such systems appears to account for almost all of the rise with redshift seen in deep surveys (figure 3).
What can have happened to these star-forming peculiars so that they were so numerous 5 billion years ago but virtually absent by the present epoch? Two hypotheses are popular. The first suggests that the peculiars are transformed via mergers or by other means into regular ellipticals and spirals. This “recent-merger” view is consistent with the predictions of hierarchical models. The second and more interesting idea is that the distant irregulars are being seen during an unusually active period in their activity, perhaps even at the moment of their formation; starved of further infalling gas, these galaxies subsequently fade to low surface brightness systems which are difficult to detect today.
As discussed above, initially there was some concern that the abundant population of faint peculiar galaxies represented misclassified spirals or spirals rendered peculiar by a short-term burst of star formation. Infrared data recently acquired with UKIRT and the HST NICMOS camera demonstrates that these optically peculiar systems appear irregular at longer wavelengths as well and, importantly, that their established stellar mass is quite significant. Although blue light could fade quite quickly if the star formation rate stopped abruptly, their infrared luminosities would hardly change. Inevitably therefore, the intermediate redshift irregulars must be represented in the census of infrared-luminous galaxies which today are dominated by regular systems. It seems, therefore, that this remarkable population must have transformed into spirals and ellipticals.
Mergers and field ellipticals
The growing evidence that mergers shaped some fraction of the Hubble sequence raises the question of whether we can quantify the rate of merging observed at faint limits and reconcile this with that expected in hierarchical pictures. Together with our colleagues in France, Canada and Australia, the Cambridge team has examined many deep fields taken with HST and conducted a census of those systems likely to merge. Around each galaxy we can, knowing its redshift, search for companions to a fixed luminosity limit and within 20 kiloparsecs. Projected line-of-sight pairs can be removed statistically. This study revealed, for the first time, a dramatic rise with redshift in the number of likely associated pairs (figure 3). Broadly speaking, the number of close pairs, corrected for known biases, increases with redshift z as (1 + z)3. However, in order to convert the rate of physically associated pairs into the true rate at which galaxies are merging, we need to know precisely the time it takes for a pair to coalesce. Whilst merging is clearly occurring, and at a rate that steeply increases with redshift, without some dynamical data on close pairs, it is difficult to make direct comparisons with hierarchical models.
We decided to consider a second approach. figure 4 shows the internal colour distributions for a selection of field elliptical galaxies of known redshift in both Hubble Deep Fields. A surprising fraction have significantly bluer cores, suggesting recently arrived younger stars. A similar study applied to ellipticals in dense, rich clusters finds far fewer examples. This difference is expected if clustered regions represent peaks in the early density field whose assembly history was accelerated. In other words, one would have to go to much earlier times in clustered regions to find similar examples.
It is natural to ask whether the decline in the number of blue irregulars is consistent with the frequency with which we see blue clumps in the emerging population of field ellipticals. Recently we have begun to examine the HDF ellipticals illustrated in figure 4 to determine the rate at which stellar mass is arriving as a function of redshift. This can then be directly compared with the predictions of hierarchical merging as well as the possibly associated demise in the population of faint, blue peculiars.
We can consider this question in terms of population statistics as well as through detailed studies of individual galaxies. Returning to the question of the infrared luminosity distributions which constrain the assembled stellar masses in various types of galaxies, at the most fundamental level if blue light from irregulars is being reapportioned in field ellipticals, we can expect the infrared luminosity density of the former to decline with time at the expense of a rise in that of the latter. Reassuringly, with some uncertainties, this transformation is indeed seen (figure 4). In the more detailed study, we can ask whether the mass associated with the “blue lumps” in the field ellipticals is consistent with an appropriate growth rate. Unfortunately, powerful though such a constraint would be, to convert the blue light seen in the HDF ellipticals into a mass arrival rate we need more astrophysical information. Are we catching the blue light during the most active stage of star formation (in which case the mass involved could be quite modest) or, more likely, is there some duty cycle of activity in which case we will be observing galaxies at various stages in an extended process. Given that many examples would be seen away from their (presumed brief) most active moment, the mass arrival rate could be more significant (figure 4).
In order to disentangle these possibilities, spectroscopic diagnostics are necessary to constrain the ages and metallicities of the young stars involved. Such programmes are ideally suited to integral field spectroscopy with the new generation of telescopes, including Gemini.
Bars and bulges in normal spirals
Any self-respecting theory of galaxy formation must also explain the more detailed structures seen within galaxies, most notably the emergence of spiral arms, the significance of the bulge or spheroidal component of galaxies and stellar bars. Modelling the internal characteristics of galaxies in the context of dark-matter theories is a less mature field, but several key predictions have emerged from theorists in recent years and many of these are directly testable with HST datasets.
As we have seen, the gradual cooling of gas into dark-matter halos leads naturally to disk-like systems, a fraction of which may later merge to form ellipticals. If these disks assemble too early they may lose their dynamical rotation in the inner regions, depositing angular momentum into their more extensive dark halos. Moreover, early formation of spirals may lead to an overproduction of present-day ellipticals since merging is much easier at high redshift. For both reasons, theorists prefer to delay gas cooling until quite recently and hope observers will find evidence for strong evolution in the numbers and sizes of spiral disks to redshifts of order unity.
Examples of “grand design spirals”, those with magnificent sweeping arms, can be found to redshifts beyond 1 and straightforward count analyses as a function of apparent magnitude and redshift detect no obvious decline in the numbers of large disk galaxies to limits where they can be recognized (figure 5). If anything, there are more disk galaxies at faint limits than expected. This appears to be because, statistically, individual spiral disks have higher surface brightnesses at high redshift with bluer rest-frame colours as expected if their star formation at that time was more vigorous. Dynamical data obtained by Nicole Vogt (also at the Institute of Astronomy) and her colleagues using the 10 metre Keck telescope, can be used to track possible evolution in the relationship between total mass (as defined by the stellar rotation in the disk) and luminosity (the so-called Tully-Fisher relation). Again there is no striking evolution in this relationship out to redshifts of 1. Whilst the samples are still small, these simple observational tests suggest that the bulk of the disk galaxy population was already in place with its present-day properties at z = 1.
However, recently we have been examining some of the finer details. A test within reach is to try and identify possible changes in the sizes of disk galaxies. The light distribution in nearby disk galaxies follows an exponential dependence on radius and the scale length of this exponential fall-off can be determined by fitting profiles to the images of the larger systems out to redshifts of 1. Early work with our Canadian and French colleagues indicated no substantial decline in the number of large disks over this redshift range. However, that study was based on only a handful of galaxies. A more precise analysis using a larger sample, based on the infrared luminosity of a galaxy as a robust measure of its integrated stellar mass, shows a modest decline in the degree of concentration of disk systems per unit mass as a function of redshift. Such analyses are giving the first quantitative estimates of the growth rate of stellar disks. These spirals grow inside large dark-matter halos and by combining the rotational velocity studies by Vogt and her colleagues and the stellar masses derived using infrared luminosities, we can also obtain direct estimates of how large a fraction of the total mass of a halo ends up in galactic stars. This fraction is of great interest in distinguishing between theoretical models. The dependence of this ratio on the mass of the galaxy, such as will soon be observationally possible by extending these comparisons to larger samples, will give important insight into the process of galaxy assembly.
What about the central bulges of spiral galaxies? These could form in various ways. Most astronomers believe they represent cores which suffered early collapse at high redshift, i.e. miniature primordial galaxies around which subsequent assembly occurs. However, they could also represent evidence for minor mergers, e.g. as miniature ellipticals in a different sense, resulting from the coalescence of unequal mass progenitors. Whereas two equal mass spirals would catastrophically destroy their respective disks in a head-on collision, forming a smooth elliptical product, in a less extreme collision one spiral might survive and adopt an elliptical-like bulge. In this respect the dark-matter picture predicts a complex interplay between types. For example, a small elliptical (or bulge) might form from two spirals or irregulars and then grow via accretion of gas into a spiral. Regardless of these continuing transformations, in hierarchical case one would expect to find most bulges statistically older than ellipticals since the latter can only be formed from the merger of near-equal mass systems.
Fortunately, bulges can be isolated in the HDF spirals to quite high redshift (figure 5) and thus it is possible to examine some of these predictions. Comparing the colours of bulges and ellipticals at the same redshift is a particularly revealing experiment since, regardless of the particular dark-matter model, a central prediction is that ellipticals should have formed more recently and be slightly bluer than bulges. In fact, we find a significant colour dispersion for bulges with many much bluer at intermediate redshifts than their elliptical counterparts (figure 5). This suggests a more recent origin for at least some bulges unless older structures are perhaps being rejuvenated by accretion of cool gas.
Bulges could also form from instabilities that arise naturally in stellar disks. Numerical simulations suggest a stellar disk can become unstable and form an inner bar which, ultimately, itself evolves into a bulge-like structure. Bars are hard to see reliably in faint galaxies; their visibility depends on the angular size and orientation of the host galaxy as well as the wavelength of observation. Notwithstanding these difficulties, we recently demonstrated (with Michael Merrifield at the University of Nottingham) that HST should be able to recognize some barred galaxies beyond a redshift of 0.6, whereas in fact we had a hard time finding any. Although bars may be transient features, the absence of any examples beyond a modest redshift makes it difficult to believe they represent a frequent route to the production of bulges which are common at all redshifts.
Where next - resolved spectroscopy?
Until recently, studies of faint galaxies have concentrated on the statistics of their integrated properties. Morphologists like ourselves have been criticized for dealing with visual features that may not have physical significance. But as we have seen, galaxy formation is most likely to be a continuous process characterized by some or all of merging, gaseous infall, dynamical instabilities and star formation timescales. It is important to move away from broad-brush pictures dominated by simpler ldquo;counting tests” and tackle the detailed astrophysics necessary to understand the present diverse population.
This hoped-for change in philosophy coincides with a revolution in ground-based capability from the new generation of actively controlled, large optical and infrared telescopes equipped with spectrographs capable of undertaking detailed internal studies of galaxies such as those displayed in this article. The Keck telescope has already shown the way via the first rotation curves discussed earlier. Resolved spectroscopy can be combined with HST mul-ticolour data to reveal star formation histories and dynamical, chemical and excitation properties for suitable sub-components. The full panoply of techniques used so successfully on local galaxies can be used, within constraints defined by telescope time and photon limits, on intermediate redshift populations.
In practical terms, 2-D spectroscopy can be achieved via mapping with a long slit or, more economically, via an integral field unit – a device which redistributes an array of spatial elements so that the spectrum of each element can be assembled conveniently on a 2-D detector (figure 6). Both the optical and near-infrared spectrographs on the Gemini telescopes (respectively GMOS and CIRPASS –) will offer such a capability (see also Allington-Smith Allington-Smith , Mountain , Parry ). Integral field spectroscopy will be extraordinarily demanding in telescope time because, unlike conventional spectroscopy, the high signal to noise essential for detailed astrophysical studies must be obtained for each sub-component of the galaxy. Nonetheless, the rewards will be great.
Resolved spectroscopy will complete a historical trend of increasing detail in the study of faint galaxies. The faint blue population was first identified in the 1970s using photography at the prime focus of 4 m telescopes. Subsequently, redshifts were added using multi-object spectrographs in the 1980s. In the 1990s HST's exquisite image quality brought resolved images, multicolour photometry and galaxy morphologies. In each case it has taken several years for observers (and theorists) to understand how to adapt to each new dimension of data. Significantly, however, each new dimension has transformed the field. We are poised now, with Gemini, VLT, Keck and Subaru, to witness a further transformation via the deployment of integral field spectrographs.
• The work of the Cambridge group has appeared in the following recent publications: