2. WHAT IS A GALAXY?

While the word ``galaxy'' still tends to conjure images of well defined spirals like M51 or giant ellipticals like M87, today's world of galaxies appears to encompass an ever widening variety of possibilities. We see galaxies spanning many orders of magnitude in both mass and energy, from systems like dwarf ellipticals, barely distinguishable from globular clusters, to distant quasars around which faint galaxian structure can sometimes be discerned (e.g. Smith et al. 1986, Yee 1992). There are galaxies with active nuclei, Seyferts, Blazars, or Extra-Galactic Radio Sources (EGRS), as well as galaxies experiencing powerful starbursts. An assortment of galaxies shows dramatic evidence for outflows, from nearby systems with bipolar X-ray plumes to classical EGRS displaying jets and lobes. Still others appear to be accreting gas from enormous X-ray ``cooling flow'' halos. And there are galaxies so distorted and convoluted by interactions or mergers as to defy classification altogether. A gallery of such galaxies is shown in Fig. 1.

(a) (b)

(c) (d)

(e) (f)

In spite of this expanding observational variety, the most enduring fundamental classification system used today is essentially one which was introduced some 70 years ago. Edwin Hubble (1925, 1926), ordered the galaxies on the basis of their optical morphology, resulting in what is today known as the ``tuning fork diagram''. The present-day identification of a galaxy as an early (late) type depending on how far to the left (right) it lies on this diagram is an artifact of historical (and erroneous) ideas that evolution is driven to the right as galaxies gradually flatten from spheres into disks and then develop spiral arms.<sup>1</sup> While more sophisticated systems have been developed over the years, for example, the concept of the ``classification volume'' introduced by de Vaucouleurs (1959) (see also de Vaucouleurs et al. 1991), or the ``luminosity'' or ``anaemic / gas rich'' classifications of van den Bergh (1960a, 1960b, 1976), the basic framework of galaxy classification has not changed significantly. The relevant question is still, ``What is the Hubble type?''.

2.2 The Limitations of Conventional Classification Schemes

Although the optical regime occupies only a small fraction of the entire electromagnetic spectrum, there is some justification, aside from the response of the human eye, for an optically-based classification scheme. This is because most (perhaps 90%) of the radiating mass in galaxies is locked up in stars which emit primarily by thermal processes in the optical regime. The Hubble sequence for spirals, in particular, does appear to reflect an ordering of physical properties (see the review by Roberts & Haynes 1994), for example, the bulge-to-disk ratio, the ratio of kinetic energy in the rotational and random components, the neutral gas fraction, star formation rate, and colour. However, other properties of spirals, some of which are quite fundamental, do not correlate at all, for example, the total mass and luminosity, the number and character of the spiral arms, or the infra-red (IR) luminosity (see Devereux & Young 1991). Elliptical galaxies present more problems still, since the Hubble type for ellipticals does not seem to correlate with any physical properties (see Djorgovski 1992). Indeed, the current debate over the relative importance of initial conditions or galaxy-galaxy interactions in elliptical galaxy formation and evolution (see reviews by de Zeeuw & Franx 1991, Barnes & Hernquist 1992a, and Section 5) casts doubt on the usefulness of a unique classification system.

Numerous categories have been introduced which do not attempt to be comprehensive classification schemes, but are rather groupings of galaxies based on specific observational traits. Examples are the peculiar morphologies of the Arp galaxies (Arp 1966), the Fanaroff-Riley jet/lobe classification for EGRS (Fanaroff & Riley (1974)), Markarian galaxies showing ultraviolet continua (see Mazzarella & Balzano 1986 and references therein), Narrow Line X-ray Galaxies (see Lawrence 1987 and references therein), IR-bright galaxies, and a host of others. This growing list reflects the modern shift away from optical observations alone to a more bolometric approach. To what extent do these categories indicate the need for a major revision of current classification systems, and to what extent are they simply minor observational perturbations on an otherwise fundamental and adequate scheme? The answer is not entirely clear. However, in addition to the limitations already mentioned, I would like to provide three illustrations, below, which collectively suggest that major revisions in galaxy classification may be required.

The first concerns objects for which the observational properties are (mostly) known, for example, the classical EGRS (e.g. Cygnus A, Fig. 1a) which have been studied for some time, at least over the optical - radio regime. If our eyes, instruments, and atmosphere were equally responsive to all wavelengths and we were developing a classification scheme based on morphological characteristics, then it is likely that EGRS would be classified according to their radio properties. The radio band contains the most ``distinguishing'' morphological characteristics and, in contrast to the optical, has an ``un-relaxed'' morphology, suggesting that this band contains much information about the source. Upon examining the spectral energy distribution, this choice would be confirmed, since the radio luminosity of such objects can rival or exceed the optical luminosity. Though the optically emitting mass would eventually be found to dominate the radio emitting mass, this is a derived, rather than an observed quantity, and would not initially be involved in the classification process. While EGRS currently constitute only a small fraction of all galaxies, their relative importance increases with redshift, z, where they likely represent a significant fraction of all massive galaxies (Peacock 1991). Carrying the argument to extremes, consider a galaxy like NGC 1275 which is at the heart of a cooling flow in the Perseus Cluster (Fig. 1f). From an X-ray point of view, NGC 1275 might be described as a giant, hot gaseous ``galaxy'' of Mpc dimension and X-ray luminosity 10<sup>45</sup> erg s<sup>-1</sup> (Schwarz et al. 1992) with a dense, optically emitting filamentary ``core'' (the current optical galaxy) and several other optical ``condensations'' (the nearby companions) throughout. Again, cooling flow galaxies are in the minority, but the list of exceptions is growing as more and better data across the spectrum become available. Indeed, there is some move now to characterize galaxies by ``bolometric Hubble type'' based on standardized images over currently available wavebands (Madore 1993).

The second refers to the question of whether we even know the important observational properties of galaxies. An example is the ultraluminous IR galaxies which radiate most of their energy in the infra-red (Sanders et al. 1988), in some cases with output rivalling the luminosities of quasars (see review by Young & Scoville 1991). These are objects which had previously been recognized as galaxies but which had not been identified as a group until their impressive IR properties were revealed by the Infrared Astronomical Satellite (IRAS), something which has occurred only within the past 10 years. That is, as recently as 10 years ago, the bulk of the energy lay hidden from view. On the opposite end of the spectrum, there are other cases in which important galaxian properties continue to be revealed, again because of improving technology. For example, more Seyfert 1 galaxies are found per unit volume in the X-ray regime than in any other waveband, and the surface density of (0.2-2 keV) X-ray emitting quasars also rivals or dominates that found in optical surveys (see review by Mushotzky, Done, & Pounds 1993). It is not a difficult extrapolation, then, to suggest that other defining properties of galaxies may yet be revealed as new and more sensitive observational windows become accessible.

The third refers to the completeness of our sample. At high redshifts, only the properties of the brighter galaxies are known because magnitude limited observations can sample only the bright end of the galaxy luminosity function (the Malmquist bias). What we are now finding, however, is that even in the nearby universe, galaxy parameter space has not yet been fully explored. Fainter observational limits are uncovering entirely new populations of galaxies previously missed because of their low surface brightness (LSB). LSB galaxies (Schombert & Bothun 1988; Schombert et al. 1992; Bothun et al. 1993; Van der Hulst et al. 1993; see Fig. 1c) appear to span a similar range of mass as normal spirals but with distinctive properties, e.g. significantly lower surface brightnesses and low star formation rates, yet rather blue colours, and they do not fit within the Hubble sequence. It has been suggested (McGaugh 1994) that they are a nearby analogue of the population of faint blue galaxies observed at intermediate redshifts.

These illustrations indicate that we are still in the data collection and classification stage of our understanding of galaxies. While incompleteness must not prevent us from attempting to make sense of the data in hand, all such attempts must be made in the context of a growing and evolving data base. Conventional classification schemes may be useful for studying the properties of ``normal'' bright spirals, but increasingly the data are outgrowing these schemes.

2.3 The Physical Properties of Galaxies

The intention of any classification system is, of course, to identify the important physical properties of the group and to understand something of the origin and evolution of the member objects, a process which has worked extremely well in the past (e.g. the H-R diagram for stars). Recent attempts have been made to do this, apart from conventional classification. Elliptical galaxies, for example, which are most poorly described within the Hubble classification system, can instead be plotted within a three dimensional parameter space defined by axes depicting mass (or luminosity, radius), temperature (or maximum velocity, velocity dispersion), and mass density (or surface density) (see Djorgovski 1992). In this space, most giant ellipticals fall within a two dimensional manifold or ``fundamental plane'', suggesting that a limited number of important processes were involved in their formation and evolution. The various projections of this plane reduce to more familiar relationships in 2-D space, i.e. Kormendy's R-µ relation (Kormendy 1977, 1980), the Faber-Jackson relation (Faber & Jackson 1976) and the cooling diagram. The fundamental plane does not represent the entire manifold, since other parameters, like the radial profile shape, ellipticity, isophotal twist rate, the presence of ``boxy'' or ``disky'' isophotes, etc. are not included and do not appear to correlate with other properties. Additional complexities also exist in that there may be slight differences between the fundamental planes for cluster and field ellipticals, dwarf spheroidals fall on a different fundamental plane, and brightest cluster members appear to obey different scaling laws than normal ellipticals. Spiral galaxies fall on a manifold within a different (2-D) space altogether, being delineated by the axes, ``form'' (B/D ratio, colour) which quantifies the Hubble sequence, and ``scale'' (L, R) which characterizes the range of luminosity over a given Hubble type. Again, there are limitations, for example, the number and character of the spiral arms is not represented, two dimensional space increases to three if the far-IR data are included, and the applicability of this scheme to LSB spirals is not yet clear. For more discussion of galaxy manifolds, see Djorgovski (1992).

Another approach is to ``teach'' artificial neural networks (ANNs) how to classify (e.g. Odewahn et al. 1992, Storrie-Lombardi et al. 1992, Naim 1994) by executing a mapping from ``galaxy parameter space'' into ``classification space''. While comparisons with existing classification schemes are still required during the training process, this technique permits a wide array of possible input parameters (e.g. < B - R >, µ₀) and has the potential of producing a more physically based, uniform, and faster framework within which classification can proceed or be adapted as required.

The physical classification of galaxies is still a developing science and, as yet, rather rudimentary compared to the relatively clean stellar evolutionary tracks of the H-R diagram, though progress is being made (e.g. Fig. 2). The difficulty is, of course, that galaxies are not ``things'', but rather collections of many different things, each sharing the same gravitational potential, evolving according to their own internal timescales and, for some components, interacting with each other. This complicates enormously the study of the evolution of isolated galaxies. But a further complexity lies in the fact that, dynamically, many (perhaps most) galaxies are not isolated. For example, the mean separation to radius ratio for stars in the solar neighbourhood is of order 10<sup>7</sup> / 1, whereas for galaxies, this ratio may be typically 10<sup>1->2</sup> / 1 (and smaller in the past). This fact has significant implications for the evolution of galaxies (see Section 5) and may help to explain why a galaxian ``H-R diagram'' still eludes us.

Figure 2. A modern view of galaxy taxonomy (from Djorgovski 1992, copyright Kluwer Academic Publishers, reproduced with permission).