Galaxy Collisions

Curtis Struck

Dept of Physics and Astronomy, Iowa State University, Ames, IA 50010 USA

E-mail: curt@iastate.edu


1. Introduction to Colliding Galaxies.

1.1 Overview

Galaxies were once seen as isolated, mostly unevolving island universes of a few characteristic types, and so rather like classical crystal structures. During the last few decades our view of galaxies has changed drastically. We now believe evolutionary processes are important even in isolated, undisturbed galaxies, and furthermore, that most galaxies are strongly, even violently affected by their environment. Specifically, as will be explained below, it is now thought that most galaxies experience several collisions or tidal interactions over the course of their lifetime which are strong enough to profoundly alter their structure and accelerate evolutionary processes (see Figure 1). Thus, collisions and interactions are now generally believed to be one of the primary drivers of galaxy evolution. The processes of galaxy formation and evolution are intimately connected to star formation, and thus, to a variety of other problems of general interest, including: the buildup of heavy elements in the universe, the formation of planetary systems, and the production and distribution of galactic cosmic rays. Hence, subjects as diverse as the solar abundance of carbon, and the rate of biological mutations in "island universes" like our own depend to some degree on the large-scale environment and collisional history of the galaxy.

For the moment, however, let us retreat to the realm of extragalactic astronomy. There is another aspect of galaxy collisions, that has not been considered much until recently. This is the use of galaxy collisions as a probe of galactic structure and stability. In brief, because of the universality of galaxy mass and kinematic distributions, it appears that the major structural components of galaxies are individually and collectively in quasi-equilibrium states. Secular evolutionary processes prevent the achievement of complete equilibrium. Moreover, the star formation rate and the distribution of the interstellar gas may be the result of dynamic self-regulation, rather than thermodynamic equilibrium. In any case, it is generally difficult to learn about the nature of an equilibrium state simply because it is a single state. The usual way to probe such variables in a dynamical system is by studying the response of the system to a perturbation. As we will see in detail below, collisions are the natural disturbances to quasi-equilibrium galaxies and galaxy disks in particular, and nature provides an abundant variety of them. This author believes that the study of collisionally disturbed systems will become the primary means of learning about the physics of star formation which is orchestrated on large-scales, and of the phase balance and other thermo-hydrodynamical characteristics of the interstellar gas.

This article is intended to provide an overview for students new to the subject and nonspecialists from other areas of physical science who would like an introduction to it. I emphasize simple physical descriptions of the phenomena wherever possible, and the physical relationships among different phenomena. In many complex situations where this is not possible, published numerical simulations give us a view of the dynamics, and frequently also provide new insights. In some cases, the theory is not yet sufficiently well developed to allow a good model to be constructed, and it is not yet possible to describe a complete dynamical theory of galaxy collisions. Nonetheless, it appears that there is consensus on many general characteristics, as well as on many of the specifics, of such a theory. Therefore, the bulk of this article is structured along a path parallel to the generic dynamical histories of collisional galaxies. The last sections consider collisions in broader environments, but the development of these environments is another closely related temporal sequence.

I will give relatively little attention to details of current numerical modeling and data analysis techniques, since these are emphasized in a number of excellent sources listed later. On the other hand, in a few sections I there is a good deal of specialist detail. Newcomers who are not especially interested in such detail will generally find overview and summary material at the beginning and end of each major section. This first chapter consists mostly of a brief historical introduction, which addresses the questions - why does anyone study a topic so remote from everyday existence, and how did these studies get started?

The literature on colliding and interacting galaxies has become too vast to fairly summarize in a single article. Whenever possible this will be a review of reviews. That is, if an up-to-date review exists already I will generally refer the reader there, and limit the discussion here. Inevitably, this means that important papers will be absent from the reference list, a circumstance that is both regrettable and necessary.

There are several excellent books or review articles that each cover many aspects of this subject, including the conference proceedings edited by Sulentic, Keel, and Telesco (1990), Wielen (1990), and Combes and Athanassoula (1993), and the articles of Schweizer (1986), Noguchi (1990), and Barnes and Hernquist (1992). Two additional publications, containing numerous technical reviews were about to become available as this article was being completed (Friedli, Martinet, & Pfenniger 1998, Sanders 1998). The author has had access to preprint versions of some of these articles, but not to the books as a whole. There are also popular books with some coverage of the subject, including Ferris (1980), Parker (1990), and Malin (1993). Parker's book, in particular, offers an interesting historical summary.

1.2 Orders of Magnitude.

Before proceeding, we should define the term "galaxy collision", which has so far been used quite loosely. In fact, it will used in a very general sense in this article to indicate any close encounter that has a significant effect on one of the galaxies involved. The term "tidal interaction" is more commonly used in the field, because the tidal gravity forces are responsible for the most significant effects. These forces are able to generate spectacular effects without involving the actual intersection of the visible parts of the two galaxies at closest approach. A near miss is as good as a hit in this field. Still, it might seem that this general definition of the term collision is misleading. However, fine distinctions are not necessary for present purposes, and in fact, may be misleading themselves. As we will see below almost all significant "interactions" involve the intersection of either the dark halos of the individual galaxies, or the mediation of a common group halo. It should also be noted that the adjective "significant" is highly context dependent, as will become clear below.

Galaxy collisions involve a tremendous amount of energy. Two objects with masses of the order of 1012 solar masses or 2 x 1042 kg meet with typical relative velocities of about 300 km/s, so the collision energy is of order 10 53 J. This is energy is equivalent to about 10 8-9 supernovae, e.g., a number of supernovae that ultimately can be produced in the merger of the two galaxies. Despite the large energy, the modest encounter velocity (about 0.1% of the speed of light) means that this is not a high energy phenomenon in the usual collisional physics sense. Nonetheless, because the energies are comparable to the binding energies of the galaxies, collisions can have very important evolutionary effects.

Galaxy collisions are extremely slow by terrestrial standards, with typical timescales of order 3 x 108 yrs., or 10 16 sec. There is little hope of observing any of the dynamics directly. Thus, when we look at the images of the hundreds of suspected collisional galaxies that are sufficiently nearby to observe in some detail what we are seeing is a snapshot gallery of systems characterized by a wide variety of structural and collisional parameters. Moreover, these systems are caught at random times in the interaction. This fact is one of the main reasons why it is so difficult to interpret the observations, and arrange the systems in a physical classification scheme. This fact reappears in many different guises below.

We shall see below that much of the collision energy is redistributed or dissipated over the interaction timescale. The dissipation rate is of order L = E/t = 1037 W. This is about the peak luminosity of one (bright) supernova. It is somewhat less than the output of a typical starburst resulting from the collision, and less than the luminosity of most quasars and other active galactic nuclei.

One of the most fascinating aspects of galaxy collisions is the fact that most of the matter involved doesn't collide with anything. In the first place, most of the mass in a typical galaxy consists of collisionless dark matter. Thus, dark matter from the companion galaxy passes through that of the target with no effects except for those due to their collective gravitational forces. Similarly, there is only a very small probablility for direct star-star collisions. The cross section of a star like the Sun is about 1017 m2, while the surface density of stars near the Sun is of order 10 per light year squared (10-32 m-2 ). This implies that the collision probability is of order 10 -15 for a typical star. The stellar density is much greater in the centers of galaxies, but the basic point is not changed.

On the other hand, the warm neutral components of the interstellar gas in the disk of our galaxy have a large filling factor (e.g. Dickey and Lockman 1990, Combes 1991). A similar conclusion is implied by the fact that the surface area of low density holes in the neutral hydrogen gas is small in other late-type galaxy disks (Brinks 1990). Moreover, the filling factor of the hot coronal or halo gas surrounding the thin cold disk in these galaxies is probably essentially unity (see e.g. McKee 1993). Thus, there must be direct collisions between the various gas components when two gas-rich disk galaxies collide. The nature of collisions between gas elements in the two galaxies depends a great deal on their thermal state. Collisions between cold clouds will be highly supersonic, e.g. characterized by Mach numbers of order 300 for clouds with a mean temperature of 100K colliding with a relative velocity of 300 km/s. At the other extreme, the sound speed in the coronal gas is of order 100 km/s, so collisions between gas haloes at such velocities will be transonic. The tidal forces in galaxy interactions which do not include direct collisions may drive waves at supersonic velocities relative to the cold gas within the disk, but the effects are less extreme.

1.3 Background and Early History

The Milky Way and the Magellanic Clouds may have experienced a tidal encounter within the last 109 years (e.g. Wayte 1991 for a brief review). Thus, there are naked eye colliding galaxies, though the effects on the Milky Way cannot be observed by simply stepping out into the backyard. One of the first "spiral nebulae" discovered with the telescope, the relatively nearby M51 system (albeit at a distance of about 9 megaparsec or 2.8 x 107 light years) has also been shown to be a collisional system. (See Byrd and Salo 1995 for a review of current thoughts on the nature of the collision). The serious scientific study of galaxy collisions began in the wake of the early efforts to discover a morphological classification system for galaxies, and the great accumulation of imagery that resulted from these efforts. I will not attempt a thorough account of the discovery of galaxy collisions here, though I think it would make a very fascinating subject for a trained science historian. However, I cannot resist some anecdotal sampling of the history, especially since it sets the scene for later developments (also see Parker 1990).

The work of Hubble and his colleagues in the first half of this century had two primary motivations. The first and best known was the desire to see whether the form of the relation describing the expansion of the universe, now called the Hubble Law, varied at greater distances and lookback times. The second motivation, which Hubble was interested in from an equally early date, focussed on finding prototypes of important classes of galaxies, with the ultimate hope that, as in stellar astronomy and many other sciences, evolutionary connections between classes would become apparent. A major result of this work is well-known Hubble tuning fork scheme, which is essentially the periodic table of extragalactic astronomy. This system is reviewed in every elementary textbook, and described in detail in The Hubble Atlas of Galaxies (Sandage 1961, Sandage and Bedke 1994).

The tuning fork diagram from Hubble's book (1958, originally 1936) is reproduced in Figure 2. The handle of the fork consists of increasingly flattened elliptical galaxies, which are dominated by old stars and have little gas, dust, and young stars. The tongs of the fork consist of two parallel sequences of spiral or disk galaxies, one with a stellar bar component, one without. In each disk sequence the prominance of the stellar bulge component relative to the star-plus-gas disk decreases to the right (e.g., from Sa to Sc). Generally, the gas fraction and young star population increases from Sa to Sc. Galaxies in the transition class S0 and the Sa class are called early-type galaxies, while those in the Sc class are called late-type. Spiral arms tend to be more tightly wound about the center in the early types and more open, but also more irregular or flocculent in the late types. This capsule description does not represent either the original classification criteria, or the modern understanding of these galaxies very well, but it is sufficient for present purposes. (See the review of Roberts and Haynes 1994 for a modern understanding.)

In the classification studies galaxies were discovered that did not fit into any of the major categories (e.g., any of the Hubble types), and whose morphologies were unusual, weird, or "peculiar". In later years it was demonstrated that while the "peculiar" bin contained a wide variety of objects, it included a number of galaxies distorted by collisions. It is interesting to look at how some examples, most of which are now very well-known and well-studied collisional systems, were described at the time.

It seems that Hubble put a number of these galaxies in his "Irregular" category, though he describes them as "highly peculiar objects" (1958, p. 47). As examples he cites NGC 5363, NGC 1275, and M82. He further suggests that "Almost all of them require individual consideration but, in view of their very limited numbers, they can be neglected in preliminary surveys of nebular forms" (1958, p. 48).

Another early general description of the classification program is provided by Shapley's book Galaxies (1943, also revised edition 1961), and there we find a little section entitled "Remark on Freaks". Shapley states,

There are also plate spirals... and frankly "pathological" types, (as Baade calls such freaks) like NGC 5128...and the ring-tail system, NGC 4038-9, shown in Figure 97.

The theories that sufficiently explain the relatively simple looking Sc spiral, like Messier 33, and the most common galaxies in Virgo, must have sufficient flexibility to take care of these aberrant types. The interpreter may need to resort to the assuming of collisions to find satisfactory causes. He will find some justification, because the individual galaxies are not so far separated but that encounters may have been fairly numerous, if the time scale has been long enough....We are only at the threshold to the house of galactic knowledge, and within there are doubtless many dark and difficult rooms to explore and set in order.

Shapley's point about the frequency of galaxy companions was echoed by Baade -

Hubble and I had a long-standing bet of $20 for the one who could first convince the other that a system which he found was single. We could never decide the bet; neither of us could pull out some distant fellow - in some cases there really was a companion and in other cases there could be. So single galaxies may be rare. (Baade 1963)

Shapley's comment about collisions probably wasn't a random speculation. It is likely that he was aware of Holmberg's (1941) article. This is evidently the first paper to present models of galaxy interactions. Holmberg's technique was to use essentially an analog computer consisting of light bulbs and photocells. The 1/r2 falloff of light intensity was to represent gravitational forces. In modern terms this was equivalent to an N-body simulation with N=37 per galaxy (74 total), and crude time differencing. Nonetheless, the expected tidal deformations were confirmed. This achievement (together with Holmberg's earlier paper (1940)) can be taken as the beginning of the theory of galaxy collisions. This seems a fair assessment even though the work described in these seminal papers already had deep roots in the Scandanavian school. For example, Toomre (1977) provides a quote from Lindblad's (1926) conjectures on (gas dynamical) galaxy collisions. Zwicky (1959) also indicates that Holmberg's work carried on Lundmark's studies of multiple galaxies, which dates to around 1920.

Toomre (1977) also emphasizes Chandrasekhar's early work on dynamical friction - "tucked away in several 1943-vintage appendices to Chandrasekhar's (1942) book". However, it was some time before this work was applied to galaxy collisions (see Toomre 1977).

A decade later Spitzer & Baade (1951) extended Holmberg's work by considering the removal of interstellar gas (a process now called "stripping") in high velocity collisions, which they argued should be common in dense clusters of galaxies. The stage was set for what is argueably the seminal observational paper in this field, Baade and Minkowski's (1954) work announcing the discovery that "the radio source Cygnus A is an extragalactic object, two galaxies in actual collision." In the next few years a number of other bright radio sources were recognized as collisional systems (see review of Zwicky 1959, sec. V). Yet the full ramifications of these discoveries (including galaxy mergers and super-starbursts) would not be appreciated until the late 1970s and the 1980s. There are many reasons for this, including the immaturity of infrared detector technologies, and the startling discovery of quasars, which received a great amount of attention in the 1960s and 1970s. Ironically, the morphology of Cygnus A was subsequently shown to be largely a result of dust obscuration of an active nucleus rather than an ongoing galaxy collision. Yet, recent observations suggest a past merger (Stockton, Ridgway and Lilly 1994).

In fact, until recently the study of colliding galaxies has been a little traveled country lane even within the world of extragalactic astronomy. Many of the great names in the field in the first half of the century had contributed, but generally only as a spinoff from other efforts. (The exceptions are Holmberg and Zwicky, who devoted much effort to "multiple" galaxies.) The primary reason for this is the rarity of these morphologically peculiar galaxies. These "freaks" were not only too few to seem important, but they provided too few snapshots to enable a coherent picture of the dynamical processes to be synthesized. To a large degree this is still true, but now computer simulations can fill in the missing frames.

Many well-known collisional galaxies were discovered serendipitously, rather like dinosaur fossils, when selected areas or individual nebulae of unknown type were imaged with large telescopes. Many of the "nebulae" came from Dreyer's New General Catalog (NGC) and his later Index Catalog additions to it (see modern version of Sulentic and Tifft 1973). A systematic observational imaging program could have discovered many of the "freaks" at a much earlier date. However, no such search was performed before the Shapley-Ames photographic survey of all galaxies above a certain limiting brightness. (A first survey went to 13th magnitude, and a later partial survey to magnitude 17.6, see Shapley 1943, 1961. Work on this catalogue has been continued by de Vaucouleurs and de Vaucouleurs 1964, and later editions.) Shapley was clearly impressed with some of the forms discovered, like the "ring-tails" NGC 4027 and NGC 4038/9 (now known as the Antennae and featured on the Nov. 3, 1997 cover of Newsweek magazine). Not all of the discoveries were NGC objects, one of the relatively early discoveries was the beautiful "Cartwheel" ring galaxy discovered by Zwicky (1941). Zwicky was very interested in "interconnected" galaxies, undertook his own surveys, and made many other discoveries (see Zwicky 1959, 1961, 1971). Figure 3 provides a summary of the morphologies he studied, and a preview of systems described throughout this article.

However, with the completion of the Palomar all-sky Schmidt camera survey, which went deeper than the Shapley (deep) survey, it became possible to carry out new searches capable of discovering many "freaks". H. Arp undertook the search for peculiar galaxies in the Palomar survey, and published his now famous atlas (Arp 1966). A similar cataloging project was carried out by Vorontsov-Velyaminov and collaborators (1959, also Vorontsov-Velyaminov and Krasnogorskaya 1961, Vorontsov-Velyaminov 1977), though he was skeptical of the idea that most of these disturbed systems were the result of tidal interactions. A great many of the objects in Arp's beautiful photographic atlas are colliders, and it has provided a starting point for many subsequent studies. Once the still more sensitive southern sky survey was completed, Arp and Madore (1987) produced a southern hemisphere atlas, with many more objects. These works, especially the original Arp atlas, "launched a thousand" observational and theoretical studies, and remain invaluable resources in this field.

On the theoretical side, Holmberg's exploratory work was followed up with the first computer models. Early works included the papers of Pfleiderer and Siedentopf (1961), Pfleiderer (1963), Tashpulatov (1969, 1970), and Yabushita (1971), which are reviewed in the Introduction of Toomre and Toomre (1972, also Toomre 1974). It was the Toomre's work, which used the restricted three-body approximation to compute the effects on the orbits of disk stars in tidal interactions, that had the greatest impact. Although the Toomres noted that a number of their results were presaged in the earlier works, their work assembled all the available pieces to make a compelling case for the hypothesis that many peculiar galaxy morphologies were the result of tidal interactions. Other papers, including Wright (1972), Clutton-Brock (1972), and Eneev, Kozlov, and Sunyaev (1973), also presented similar, and confirming, numerical results. Most of these projects seem to have begun independently, though Wright acknowledges communications from the Toomres in his paper. Another important strand of this fabric was the analytic work on the Impulse Approximation of Alladin and collaborators in the 1960s and 1970s, which will be discussed below (see references in the review of Alladin and Narasimhan 1982)

From this point up to the present the field has grown very rapidly, and expanded in many directions, making it impossible to capture all the important developments in a brief historical summary. In fact, it is impossible to present all these developments in this modest review, so what follows will be a sampling.

1.4 The Importance of Collisions.

Before delving into the details of the more recent research, however, we should explicitly state some of the motivations for this activity. As described above, it was becoming clear by the mid-1970s that many of the morphologically "peculiar" galaxies, i.e. those that didn't fit into the standard classification schemes, could be accounted for as the result of tidal interactions. But from the beginning it was clear that these galaxies are rare, and so we might wonder, what is their importance? To roughly estimate their "rareness" consider two catalogs based on the Palomar northern sky survey. The Zwicky (1961) catalog of "Galaxies and Clusters of Galaxies" has some 30,000 objects, while the Arp atlas has 338 interacting pairs or groups. This implies that colliding galaxies are of order 1% of all galaxies. However, this estimate is too "rough", as we will see later in this section.

Toomre and Toomre (1972) offered some, at the time speculative, suggestions on these matters that generated much interest, and, in fact, ultimately became the dominant ideas in the field. These ideas were based on, but extended well beyond, the results of their collisionless (star-like) test particle simulations. To begin with, they noted that the observed tails and plumes were successfully reproduced in models involving close collisional encounters, and that such events were unlikely to be the result of galaxies approaching on random hyperbolic orbits. They argued that such collisions were more likely to occur between galaxies on eccentric, bound orbits. They then took the argument a step further and suggested that such large-scale tidal distortions must be formed at the expense of orbital energy, so that the two galaxies must inevitably merge (see discussion and early history in Toomre 1974). A third deduction follows -

...Would not the violent mechanical agitation of a close tidal encounter - let alone an actual merger - already tend to bring deep into a galaxy a fairly sudden supply of fresh fuel in the form of interstellar material, either from its own outlying disk or by accretion from its partner? And in a previously gas-poor system or nucleus, would not the relatively mundane process of prolific star formation thereupon mimic much of the "activity" that is observed? (Toomre and Toomre 1972)

E.g., interactions and mergers may funnel interstellar gas into the central regions of galaxies, and trigger enhanced star formation. Specific mechanisms are not described in detail, though a couple are implied. First of all, there are the direct tidal effects, the "mechanical agitations" of the quote. Secondly, in the paragraphs following the quote they propose that the encounters might drive strong spiral waves as in M51. These waves could also enhance angular momentum transport and channel gas inward. As we will see below, there is now strong observational and theoretical support for these ideas about merging, fueling galactic centers, inducing prolific star formation, and the more general notion that these processes can have a profound impact on the evolution of the individual galaxies.

An important study by Larson and Tinsley (1978) provided early observational confirmation of star formation enhancements in interacting galaxies. Larson and Tinsley studied the broad band optical (UBV) colors of the Arp atlas galaxies, and compared them to those of the Hubble atlas, assuming that the "normal" galaxies of the latter could serve as a control sample. They further produced a grid of color evolution models for aging stellar populations with a variety of star formation histories. These ranged from cases with a constant star formation rate over 10 billion years (10 Gyr.), to models of populations with all their stars formed in a relatively short burst (e.g. of duration 0.02 Gyr.). The colors of the burst models, and of combination models with a significant burst component, evolve significantly in the first Gyr. after the burst. Thus, large color variations were predicted in galaxies with significant recent bursts of star formation, and indeed, they found that,

Normal galaxies have colors that are consistent with a monotonically decreasing SFR... In contrast, the peculiar galaxies have a large scatter in colors that is consistent with bursts as short as 2 x 107 yr. involving up to (about) 5% of the total mass. Nearly all of this scatter is associated with galaxies showing evidence of tidal interaction... These results provide evidence for a "burst" mode of star formation associated with violent dynamical phenomena. (Larson and Tinsley, 1978).

In the succeeding years a great deal of evidence was obtained in a wide range of wavebands to support the conclusion that collisions and interactions frequently drive a much enhanced star formation rate, though there are exceptions to the rule. We will take up this topic again in a number of sections below; also see the reviews of Keel (1991), Barnes and Hernquist (1992), Elmegreen (1992), Mirabel (1992) and Kennicutt (1998a).

In the quote above, Toomre and Toomre raise another important issue - the connection between galaxy collisions and nuclear activity in galaxies. Like the question of collisionally induced star formation, this topic has received much attention in the last couple of decades, and we will summarize this story in section 8. For the moment we merely note that such a connection has remained much more elusive than in the case of induced star formation. The question posed by Toomre and Toomre, whether extreme star formation in the central regions might mimic nuclear activity, has also been revived in recent years (e.g. Terlevich et al. 1992, Terlevich 1994, and references therein). For example, a vigorous debate developed around the question of what powered the ultraluminous, infrared galaxies discovered through the analysis of IRAS (Infrared Astronomical Satellite) data, enormous starbursts, active nuclei or both (see the review of Heckman 1990 and other papers in those proceedings). It is believed that these galaxies are primarily merger remnants, so in any case collisions were implicated.

In sum, there is now nearly overwhelming evidence from observations, and numerical models that collisions can strongly disturb the morphology and evolution of the galaxies involved, both by direct gravitational "agitation", but also indirectly by driving strong star formation. As we will see below, the latter process leads to the conversion of large quantities of gas to stars, the creation of whole new stellar populations, and massive changes in the distribution and thermal phase balance of the remaining gas. We will also see that there is now strong evidence to support the Toomres' conjecture that most collisional encounters are mere preludes to the eventual merger of the galaxies. However, for the moment we will leave the merger issue, to reexamine the second fundamental question - how uncommon are galaxy collisions? While they are rare on the sky, we now know that they profoundly affect the galaxies involved, so the relevant question is - how likely is it that a galaxy will experience a significant collision in its lifetime.

Toomre (1977) noted that in many of the systems with conspicuous tidal tails the centers of the two galaxies were very close. This fact, and the other available evidence, suggested that these galaxies were nearly merged. He provided 11 outstanding examples, and he estimated that of order 10% of the galaxies had participated in major merger sometime in their lives. Noting that this was close to the fraction of elliptical galaxies, and that induced star formation (and other processes) would tend to change a merger remnant to an earlier Hubble type than that of its predecessors, he speculated that most ellipticals might be formed from spirals in mergers. We will not be able to explore all of the huge literature that has grown up around this hypothesis in subsequent years, the reader is referred to the review of Hernquist (1993, and chapter 6). What is important for present purposes is the fact that, while on one hand this hypothesis generated a strong and long-lived debate (e.g., Ostriker 1980, Parker 1990, pgs. 198-202, and the articles in the final section of Wielen, 1990), on the other hand, it fractured the concensus that strongly interacting galaxies were very rare, and thus, unimportant.

The earlier concensus rested on the assumption that random galaxy collisions would be unlikely, since the mean distance between galaxies is large compared to their sizes. What the Toomres and others at the time discovered was that the evidence suggested that most collisions occured between galaxies in groups that were at least loosely bound. I.e., collisions were built into the initial conditions, galaxies are born in groups. So why did they appear to be rare? Because the collision timescale is less than an order of magnitude of the age of the universe (Toomre 1977, results of more recent numerical work are reviewed in Barnes and Hernquist 1992). Thus, even if collisions happened at random times we would never see more than a fraction of them. Actually, as Toomre pointed out, there are good reasons to believe collisions were much more common in the distant past. We are in an age of increasing studies of high-redshift galaxies by the Hubble Space Telescope and a new generation of ground-based telescopes, and preliminary results indicate that this is indeed the case (see several relevant articles in Benvenuti, Macchetto, and Schreier 1996 and section 9.3). Finally, we note that while Toomre focussed on the extreme case of mergers between two (equally) massive progenitors, where the disruption and dynamical heating is great enough to form an elliptical galaxy, collisions and mergers between unequal partners are probably more common. Such collisions, and eventual mergers, can still have a dramatic effect on the evolution of the larger galaxy, as we will see below.

It has also been realized in recent years that even the assumption that almost all collisions involve only two galaxies may be incorrect. Again this is not the result of chance, but of the overall collapse of (loosely) bound groups of galaxies (see chapter 9 and the recent papers of Governato, Tozzi, and Cavaliere 1996, Weil and Hernquist 1996, and references therein). Indeed, Weil (1994, Weil and Hernquist 1996) has recently discovered that numerical simulations of multiple mergers can produce remnants that more closely match the structural and kinematic details of some ellipticals than binary merger remnants.

These theoretical ideas have some profound observational consequences, most of which will be explored later in this article. First, the galaxies with extreme morphological disturbances may be only the tip of the iceberg as far as collisions are concerned, because although spectacular, this is a short lived phase. Then the question becomes, what evidences of collisions can be found later? The elliptical galaxies with faint shells or ripples around them, discovered by Malin and Carter (1980, 1983), and interpreted as tidal debris by Quinn (1982, 1984), provide one of the best examples (see section 5.5). Seitzer and Schweizer (1990) found that 32% of the S0 galaxies and 56% of the elliptical galaxies in their sample have "ripples". As a second example Scoville (1994) estimates that in the last 109 yrs. 2% of the spiral galaxies became luminous infrared galaxies as a result of a merger or strong interaction. The arguments Toomre applied to elliptical galaxies would suggest that the overall fraction of spirals experiencing such an event in their lifetime might be at least 10 times larger. (On the other hand, if most of these spirals turned into ellipticals there would be uncomfortably many, so simple-minded extrapolation may be dangerous here!) A third and final example, Odewahn (1994) finds that all but 4 in his sample of 75 Magellanic (very late type) spirals have close neighbors, so tidal interactions are likely to be very important in this class.

In sum, collisions can profoundly influence the evolution of the individual galaxies involved in a cosmologically short time. While violently disturbed galaxies appear rare on the sky, a collision may occur between one and several times in the life of a typical galaxy. There are increasingly strong theoretical motivations and observational indications that collisions are one of the most important processes in galaxy evolution.

1.5 Nature's Galaxy Experiments.

There are many aspects of galaxies about which we know relatively little, including the dynamics and thermal processes in the interstellar gas, which are almost certainly coupled over a large range of scales by turbulence (e.g., Scalo 1990). Similarly, there is a great deal to learn about the mechanisms of star formation in galaxy disks, especially large-scale, wave-driven star formation. Colliding galaxies can be viewed as nature's own experiments, ideally suited to probe structure and dynamics, in much the same way that accelerator experiments probe the micro-world. As Arp (1966) put it,

The peculiarities of the galaxies...represent perturbations, deformations, and interactions which should enable us to analyze the nature of the real galaxies which we observe and which are too remote to experiment on directly. ...From this range of experiments which nature furnishes us, then, it is our task to select and study (those) which will give the most insight into the composition and structure and the forces which govern a galaxy.

Collisional perturbations come in a range of strengths, depending on the mass and compactness of the galaxies, and the distance of closest approach (see Sec. 2). There is a perturbative limit, where a low-mass companion interacts with a massive primary. This case is especially interesting for studying interstellar gas dynamics and induced star formation, because the primary disk is disturbed, but not disrupted, in a single encounter. At the other end of the scale, there are mergers between nearly equal progenitors, which test the nonlinear stability of all components of the galaxies. The outcome of all types of collision depends on the structure of the dark matter halos, so at least statistically, the comparison of models and observation can provide information on these halos. Examples of all of these applications of the "experimental viewpoint" will be given below.

This viewpoint also comes naturally with numerical modeling, where we can do experiments on "galaxies". It should already be clear from the brief history above that the close interaction between computer modeling and observation has been the key to progress in this field, and it will be a recurring theme in the rest of this article.