Dept of Physics and Astronomy, Iowa State University, Ames, IA 50010 USA
7. Induced Star Formation
We have described a number of ways in which collisions and mergers profoundly affect the evolution of galaxies. However, one of the most important, and among the most brilliant, is induced SF. This is a daunting topic, because there is a vast literature. This ranges from detailed studies of the SF morphology in individual systems, to general theories of SF in galaxies, and how they might apply to interacting systems in particular. As to the former, we will not consider many specific systems beyond those already described. This chapter consists of a more general summary, and a look at some of the relevant mechanisms. Yet even this restricted area contains a large body of literature, and is a very fluid and rapidly changing field. Fortunately, there are a number of specialized review articles (e.g., Keel 1991), and conference proceedings that contain reviews of many relevant aspects. These include the recent volumes edited by Franco, Ferrini, and Tenorio-Tagle (1992), Tenorio-Tagle, Prieto, and Sanchez (1992), Tenorio-Tagle (1994), and Shlosman (1994). Note, we will not consider techniques for measuring SF in galaxies, and their relative advantages and disadvantages in detail. The interested reader is referred to the reviews of Kennicutt (1990, 1998a).
7.1. Color, Ha and Other Indicators of Global Enhancements
The early work of Spitzer and Baade (1951) on gas splashed out in collisions, and Zwicky's (e.g., 1959) emphasis on the importance of tidal distension, may have discouraged thinking on the possibility of gas compression and induced SF. If so, Toomre and Toomre's (1972) conjecture that collisional disequilibrium plus dissipation will lead to gas funneling into the central regions of interacting galaxies, changed the thinking. Larson and Tinsley (1978, henceforth LT) followed up on this suggestion with an extremely influential paper comparing the broad band (UBV) colors of (mostly normal) Hubble atlas galaxies to (collisional) Arp atlas galaxies. They found that the mean colors of the Arp atlas galaxies were somewhat bluer than the Hubble atlas galaxies, but that the color range of the Arp galaxies was much greater. They compared these samples to color evolution models consisting of two components: 1) an old stellar population with a continuous, but exponentially declining SFR, and 2) a relatively young population formed in a "burst" of short duration. Most of the Hubble galaxies could be accounted by models with little or no burst component, but a modest range of timescales for the declining SFR. To account for the color range in the Arp galaxies a substantial burst component was needed, as well as a range of times since the burst.
This successful fit of models to optical observations, though not unique, gave rise to some durable ideas about collisionally induced SF. The first of these is that the nature of this SF is burst-like, rather than simply an enhanced level of continuous SF. The suggested timescale for these bursts was of order 107 yrs.; much shorter than the dynamical or merger timescale, implying that SF would not be greatly enhanced throughout the collision and merger process. (Note: that the possibility of multiple bursts was one of the variables not considered in LT.) A second conclusion is that because a range of burst strengths and ages would be represented, the primary characteristic of the colors of collisional (burst) galaxies would be a large range, rather than extreme blueness.
At about the same time Huchra (1977) published a similar study of blue Haro, Zwicky and Markarian galaxies, which included similar color evolutionary modeling. He likewise concluded that most of these galaxies consisted of composite stellar populations, i.e., an old component as well as a substantial burst component. He found no "compelling evidence" for a population of purely young galaxies. Shortly thereafter Struck-Marcell and Tinsley (1978) published models for near-infared colors. Though there was little data to compare to, these models predicted even greater color variations in composite populations over the longer infrared-to-optical baseline.
In the time since these early studies, color evolutionary modeling has become a standard tool in the study of SF history in galaxies, despite the fact that the large number of possible variables often make conclusions based on it ambiguous. We will content ourselves here with citing a few relevant highlights. Many sophisticated models (e.g., the large grids of Arimoto and Yoshii 1986) were developed more for application to long term galaxy evolution and the colors of early-type galaxies, rather than for studying color changes on the shorter timescales of galaxy collisions. The models of Arimoto and Bica (1989 and references therein) described rapid changes in colors that occur with the onset of the red supergiant and red giant evolutionary phases in young star clusters, or other coeval populations. These potentially important and observable changes were not adequately represented in earlier models. The models of Charlot and Bruzual (1991, Bruzual and Charlot 1993) also included a more sophisticated treatment of the giant phases, and a new interpolation technique for using stellar evolution tracks. An extensive compilation of current models, and model inputs has recently been published by Leitherer et al. (1996).
One of the most important indicators of recent SF is the flux of the H a line. Kennicutt and others investigated the range of Ha emission in normal galaxies, and interpreted it with the aid of evolutionary models (see Gallagher, Hunter and Tutukov 1984 and the discussion and references in Kennicutt 1990, 1998a). This provided a basis for comparison to collisional galaxies in the studies of Keel et al. (1985), Bushouse (1986, 1987) and Kennicutt et al. (1987). In addition to the Ha fluxes, the first paper also provided nuclear spectrophotometry of a large sample, and the latter study included IRAS far-infrared fluxes for comparison (also see Bushouse, Lamb, and Werner 1988). The basic results echo the earlier color studies - when compared to a control sample, the interacting galaxy sample had systematically higher emission fluxes in both Ha and the far-infrared. The dispersion in the emission fluxes was also greater in the sample of interacting galaxies. And finally, the Ha fluxes could be accounted for with the addition of a short burst component in the SFR. One which produced a small (but bright) fraction of the total stellar mass of the galaxies.
From this point on the case for global SF enhancements in galaxy collisions was demonstrated to the satisfaction of most. However, there are a number of complications. An early observational one was connected with searches for enhanced radio continuum emission, which can result from either SF or nuclear activity. Because of limited resolution, it was not clear initially whether or not radio continuum enhancements in strongly interacting systems were entirely due to active nuclei (e.g., Hummel 1980, 1981). High surface brightness star-forming continuum disks were subsequently resolved, e.g., Hummel et al. (1987), and especially in the extensive work of Condon and collaborators (e.g., Condon et al. 1982, Condon and Broderick 1988, Condon et al. 1990, and Condon et al. 1996).
A number of works have also supported the idea that the strongest SF enhancements occur in the most "violently" interacting systems (e.g., Kennicutt et al. 1987). This includes the multiwaveband studies of Bushouse (1986, 1987, Bushouse et al. 1988, Bushouse and Werner 1990), which were aimed specifically at such systems. Observations of molecular (CO) emission, which is another important indicator of SF activity (see review of Young, 1990 and Young et al. 1996) have been used to estimate SF efficiency, which is found to be extremely high in merger remnants. In fact, it is now believed that a significant fraction of all stars ever created, and of the metals scattered in the intergalactic medium, may have been made in young merger remnants (see the previous chapter and refs. therein).
While the spectacular results on merger remnants captured the spotlight, younger and nonmerging collisional systems received less attention. However, there have been a few studies examining the general question of the time dependence of SF in young to intermediate aged collisional systems, in addition to "case studies" of individual systems too numerous to cite here. Bushouse's (1986, 1987) sample included a number of systems that seemed to be near the closest approach or impact point in a first collision. There are substantial enhancements in these young, violent systems. Another important work is Keel's (1993) study of "kinematic regulation". In this work Keel found that the most important variable in determining SF enhancements is the relative amplitude of the velocity disturbance. His sample consisted of galaxy pairs with low relative velocities, and with what we can call young to intermediate ages. Sample galaxies also passed a geometric selection criteria, which required that at least one member be within 30° of edge-on, and that its companion be within 30° of the projected disk plane. These criteria allowed an optical rotation curve to be obtained, and a good estimate of whether the collision was prograde or retrograde for the galaxy under study. The sense of the collisional orbit, the projected separation, and a number of other variables were found not to correlate significantly with the global SF. The velocity disturbance, which was found to be the important variable, depends on both the magnitude of the disturbance and the galaxy's susceptibility. In support of the latter dependence, enhanced SF was commonly found in galaxies with large regions of solid body rotation. Although a detailed comparison is probably not possible, these results seem in qualitative agreement with the dynamical models of induced waves and bars discussed in the preceding chapters.
7.2. Spectral Line Diagnostics
7.2.1. SF Histories
The studies described in the previous section indicate that large amplitude disturbances resulting from direct collisions or strong tidal encounters are able to generate vigorous starbursts before the galaxies merge, and super-starbursts afterward. Beyond this, colors, Ha fluxes and many of the classical SF indicators do not have the sensitivity to elucidate the duration of bursts or the history of SF following a collision. Recently, attention has focussed on using optical and infrared spectral line flux ratios to obtain additional information on these topics.
A number of authors (e.g., Bica, Alloin & Schmidt 1990) have pointed out an age versus burst strength ambiguity in interpreting the broad band colors of old-plus-young mixed populations. The rapid evolution of the most massive stars means that very blue colors are very short-lived. Less blue colors can result from either a weak burst mixed with an old population, or a strong, but aging burst. Spectral indicators like the Balmer absorption lines, calcium H and K lines (Leonardi and Rose 1996), or infrared CO bands (Bernlohr 1992, Campbell and Terlevich 1984) can be used to distinguish and date the starburst component.
Except for a few nearby systems, the application of such methods to colliding galaxies is relatively recent. Bernlohr (1992) computed a set of evolutionary models for both broad band colors and a couple of dozen spectral lines, and then applied them to the interpretation of the prototypical starburst galaxy M82. We will return to discuss that object in a moment. Bernlohr (1993) applied these models to the study of a sample of about 30 interacting systems. The spectra he obtained for this sample were inevitably of lower resolution and sensitivity than nearby galaxies, and so, there were fewer lines to fit. Nonetheless, he obtained several interesting results concerning SF timescales in interacting systems.
The first of these was that all the starbursts in his sample had ages of less than 2 x 107 yrs., from which he concludes that starburst duration is generally less than this time. Recall that Larson and Tinsley (1978, also see Kennicutt et al. 1987) had concluded that the colors of Arp atlas galaxies could be accounted for with an old population and an aging burst population, where the burst was of equally short duration and modest strength (£ 10%). Published age estimates of individual starburst galaxies have also generally been short. Specifically, there has been no evidence for enhanced SFRs on intermediate timescales, despite the fact that dynamical and gas consumption timescales are typically intermediate (e.g., 108 - 109 yrs.). The implication is that starbursts are not ended by (global) starvation.
Bernlohr notes an important caveat, however. He had no information on heavily obscured SF regions, such as those in merger remnants. In light of the discussion of ULIRGs above, we should not be surprised if the timescales and the nature of SF are very different in merger remnants.
Bernlohr also found that the minor or secondary galaxies in his sample pairs were much more likely to be in burst or post-burst phases than the primary galaxies. He estimated that time delays between bursts in the two partners range up to several times 108 yrs. Bushouse (1986) found a similar result in his violently interacting sample, but Kennicutt et al. (1987) noted a tendency for galaxies in their pairs to have comparable Ha equivalent widths. Joseph et al. (1984) found a similar result, Lutz (1990) and Telesco et al. (1988) found the opposite result in samples selected by their far-infrared emission. There are many dynamical variables, so it is not surprising that the situation is very complex. Except for special subclasses of colliding galaxies, it seems likely that this result will not be strong in large samples.
The art of spectral synthesis, and its use for deciphering SF histories, is advancing rapidly in many areas of extragalactic astronomy, but especially for young stellar populations. (The technique has also begun to be applied to data in other wavebands, e.g., P?rez-Olea and Colina's (1995) recent evolutionary models of the radio emission in starbursts.) Garcia-Vargas, Bressan and Diaz (1995 and references therein) have produced a large set of model spectra for young star clusters (age < 6 x 10 6 yr.) and their surrounding ionized gas regions. The rapid evolution and high mass loss rates of massive young stars provide some distinctive color and spectral signatures. The rapid evolution is responsible for color shifts to the red as stars evolve to the red supergiant phase. The appearance of absorption lines, notably the near-infrared triplet of ionized calcium, is also associated with this change. The heavy mass loss is responsible for removing the hydrogen envelopes, and revealing layers enriched in heavier elements, which characterize Wolf-Rayet stars. These stars are identified by helium lines, and emission lines of nitrogen and carbon in multiple ionization states (see the review of Massey 1985). The Wolf-Rayet phase immediately proceeds the red supergiant phase. Finally the emission line intensities of the surrounding gas depend on the flux of ionizing photons, and the ionizing flux per unit mass of young stars is age dependent.
Garcia-Vargas et al. have recently applied these models to the spectra of several regions in the colliding galaxy pair NGC 7714/15 (Arp 284). They argue that the spectra of at least one of these regions in the center of the primary galaxy is best fit as the combination of two mini-bursts of age 3-5 and 7-9 million years. (This system was also in Bernlohr's (1993) sample, where he classified the primary as a starburst and the secondary as in a post-starburst phase.) Although high quality spectra are required, it seems likely that many more studies of this type will be undertaken in the coming years.
7.2.2. IMF Variations and the Example of M82
Since most of the luminosity of a young burst population is due to the massive stars, it is difficult to determine the mass of the low mass stars produced in the burst, and thus, the total mass of stars formed and the importance of collisionally induced starbursts in galaxy evolution. This is especially true when optical/ultraviolet color and spectral line observations are used. In principle, infrared observations are more sensitive to the low mass stars, since not only are they brightest in the red parts of the spectrum, but in a burst population they will often be enshrouded in dusty "birth cocoons". However, in the near-infrared there is the potential for confusion with old stellar populations or red giants. More generally, these, and other problems, prevent us from determining the so-called "initial mass function" (IMF), which describes the relative number of stars formed in different mass ranges.
This is a general problem in extragalactic astronomy, not peculiar to colliding galaxies. There has been a great deal of progress in the last couple of decades in determining the IMF of star clusters in the Milky and the Magellanic Clouds, see e.g., the reviews of Scalo (1986), and Garmany (1994). It is widely believed that the function has a universal, approximately power-law form, like that first suggested by Salpeter, except possibly at the high mass end. However, this very complicated topic is beyond the scope of this review. We refer interested readers to the recent review of Hunter et al. (1997), Leitherer (1998), and especially Scalo (1998) for a critical look at the observational foundations of the universal IMF idea.
In particular, it has been suggested that IMFs may be different, or at least extend over different ranges, in galaxies with very high SFRs. That is, when the high SFRs in the objects we now call starburst galaxies were discovered, it was immediately recognized that with these SFRs the gas consumption timescales are much shorter than the age of the universe, assuming an IMF like that in the Milky Way (e.g., Sargent and Searle 1970). Thus, either the high SFR is not maintained for very long, or the IMF is different or both. The existence of a universal IMF would support the first possibility, hence "starbursts." One example of the difficulty of observationally constraining the nature of individual starbursts is the nearby starburst and collisional galaxy M82, a member of the M81 group.
In the late 1970s several papers suggested that the vigorous activity in the central regions of M82 was the result of massive, young stars recently produced at a very high rate (Solinger, Morrison and Markert 1977, O'Connell and Mangano 1978, Rieke et al. 1980). The latter paper, in particular, included a detailed stellar population model of the starburst, despite the fact that M82 is nearly edge-on and highly obscured by dust. Based on a variety of near infrared observations Rieke et al. concluded that the extinction from the nucleus of M82 was as great as AV » 25 magnitudes (a factor of 10 10 in the optical), though they suggested that it could be quite nonuniform across the central regions. Such large optical extinctions implied non-negliglible extinctions in the near-infrared as well, and that the overall starburst luminosity is much greater than observed in those bands. The substantial UV and 2 micron fluxes constrain the duration of the burst to be more than 107 yrs. The kinematically determined mass (and the gas mass), provided some of the tightest constraints on the mass of stars produced. To avoid producing too great a mass of stars, Rieke et al. argued that the burst age had to be £ 108 yrs., and the IMF had to be very deficient in stars of mass £ 3 solar masses relative to the "standard" (local group) IMF. The best-fit models predicted a high supernova rate, whose young remnants were subsequently detected by radio continuum observations (Kronberg Biermann and Schwab 1985). Far-infrared observations (Telesco and Harper 1980) also confirmed the high luminosity.
The tale of M82 took some twists with the papers of Lester et al. (1990) and Telesco et al. (1991). The first of these papers presented extensive near infrared spectroscopy of molecular and atomic hydrogen, and ionized iron. On the basis of these data the authors argued that the extinction of the infrared emitting region was much lower than previously thought (AV » 5 magnitudes), and moreover, that emission from hot dust makes a substantial contribution at wavelengths near 2 microns. These conclusions considerably modify the constraints on population models, suggesting that the burst population is dominated by red supergiants (and thus in a very short-lived phase), and that a truncated IMF is not necessary. The paper of Telesco et al. presented near and mid-infrared mapping of M82. The primary result of this paper was the discovery of a 1 kpc bar, i.e., one which extends across the starburst and molecular gas region, and which might be responsible for funneling gas inwards. In addition, the authors concluded that hot dust emission is significant at mid-infrared wavelengths, and supported the idea of a lower value for the visual extinction.
Bernlohr (1992) and Rieke et al. (1993, also McLeod et al. 1993) have new data, and produced a new generation of starburst models. Bernlohr used new spectral constraints, and an extinction free estimate of the ionizing flux from radio recombination lines (see Puxley et al. 1989, 1991). His results are generally similar to Rieke et al. (1980), except that different stellar evolutionary inputs yielded somewhat different K band (2.2 micron) fluxes in the models. The extinction issue was not addressed, beyond a statement on it's patchy nature. The papers of Rieke et al. and McLeod et al. present models that include nonuniform dust "screens" as well as homogeneous ones, and conclude that the "visual extinction to the nucleus lies between AV » 12 and AV » 27. (We also note the recent infrared spectrophotometric study of Smith et al. (1996), which found an average extinction of 25 magnitudes in 20 luminous starburst galaxies.) They argue that with patchy extinction and a new value for an H2O band strength the need for hot dust emission at 2 microns disappears. The parameters of their final best fit model are very similar to the Rieke et al. (1980) model, with an IMF truncated at the low end.
More recently Satyapal et al. (1995) presented extensive, Fabry-Perot spectral-imaging observations of near-infrared hydrogen recombination lines. With many lines measurements at high-resolution they were able to address the ambiguities of earlier studies. For example, they describe how previous results were based on flux ratios derived by scaling large aperature data to a smaller aperature, in order to compare to the small aperature flux measured in another line. On the basis of their own data they find that the visual extinction to the starburst region varies from 2-12 magitudes, and that the 2.2 micron K-band luminosity is much lower the previous estimates. Their results suggest little need for an IMF truncated at the low end, in accord with Lester et al. However, in agreement with McLeod et al. (1993), they find little hot dust emission in the K-band. The spatial variation of the Brackett g flux, "along with other star formation diagnostics, suggests that the nucleus contains later-type stellar populations, and the starburst phenomena is propagating outward." This latter result is confirmed with additional data in Satyapal et al. (1997).
One moral of this rather long story is that, even in a nearby starburst galaxy with little or no nuclear activity, it can be very difficult to decipher the stellar populations, SF history and IMF. It still seems likely that very low values of the mass-to-light ratio (e.g., M/L << 1.0, in units of the solar mass and luminosity) indicate an IMF with few low mass stars. The M/L constraint is one of the most fundamental in M82, and the debate hinges on the L, determined from the K-band. Generally, both M and L can be very difficult to determine. The frequent suggestion that interacting galaxies in general have an IMF enriched in high mass stars (e.g. Kennicutt et al. 1987, Rieke 1991) remains uncertain.
A related indicator has been studied by Young et al. (1996, Young 1993, and references therein). They have used the ratio of massive star luminosity to gas mass (L/Mg, with Mg derived from CO observations) as a measure for high mass SF efficiency, and found that while this factor varies little among different Hubble (disk) types, it is significantly elevated in interacting galaxies. How much of this enhancement is due to modified IMFs is not yet clear.
On a more optimistic note, we do seem to be converging on a good overall picture of the starburst in M82, which is the prototype. The primary starburst region is often described as a disk or ring of radius less than a 0.5 kpc. at the inner edge of an annular disk of molecular gas. However, Telesco (1988) has suggested, on the basis of the kinematic observations, that the molecular gas may located in spiral arms. He also provides us with a graphic description of a starburst region -
If much of the apparent thickness of the IR ridge is due to projection of a thin tilted disk, then the surface brightness corresponds to an average separation of only a few parsecs between Orion-like complexes, each emitting » 3 x 105 (solar luminosities)...Thus, the center of M82 is filled with OB stars, supernova remnants, and a complex, dynamically advanced distribution of overlapping HII regions permeating a relatively warm, filamentary medium bathed in UV radiation. The effect of these OB stars and supernova remnants is so profound that > 107 (solar masses) of gas are being driven from the plane of M82 (in a galactic wind). (Telesco 1988).
The (stellar) bar is also found in this region, and not only helps explain the fueling of the starburst, but if it is an induced bar, provides a connection with the large-scale interactions in the M81/M82 group. The galaxies of this group are indeed strongly interacting, as is illustrated by the large scale HI maps of Appleton, Davies, and Stephenson (1981), Appleton and van der Hulst (1988), and Yun, Ho, and Lo (1994), see Figure 25. In fact, it appears to be a complex group interaction, involving several galaxies or multiple encounters (e.g., Thomasson and Donner and references. therein).
7.3. SF Region Morphologies
In fact, we have already described many of the important morphologies of induced SF regions in sections 2-6 above. However, most of those discussions were in the context of collision dynamics, with the SF serving primarily as an illuminator of characteristic morphologies and an indicator of gas compression zones. Nonetheless, to summarize, in Table 2 we list some general types of SF induced by galaxy collisions, give some examples, and provide an index to where they are discussed in this paper. The first three types can be induced in non-merging collisions. Mergers themselves can induce all three types, and potentially with much greater intensity, depending on the gas contents and relative sizes of the collision partners (see e.g., Schweizer 1990, 1998, and Barnes and Hernquist 1992a). Table 2 also catalogs three general properties of these morphologies: location, timescale and relative intensity. The latter two are qualitative generalizations, since there are large variations between systems, and except for a few well-studied systems, our knowledge is very incomplete.
The characterization of the intensities is based on gas consumption times or SFRs, which can be estimated from far infrared luminosities, albeit with substantial uncertainties. For example, the typical formation rates of stars of mass greater than 2.0 solar in normal spiral galaxies are about 3 solar masses per year or less (e.g., Solomon and Sage 1988, Kennicutt 1990, Young et al. 1996 and references therein). Collisional systems like the Cartwheel (Higdon 1996) and M82 (McLeod et al. 1993) have moderately enhanced rates of a few to 10 solar masses per year. Merger remnants range upwards from a few tens of solar masses per year (Solomon and Sage 1988). Thus, if SF continues for times of order 3 x 108 - 10 9 years in merger remnants, which is probable with continuing gas inflow.
The lower SFRs and shorter burst durations in collisional systems with central starbursts, like M82, suggest that the net SF in these systems is a small fraction of the whole. Collisionally driven waves also have relatively low net SFRs, but depending on the type of wave, have a wide range of durations. For example, ocular features are very short-lived (see section 3.5), as are compression regions associated with high order caustic waves (section 3.3). The circular waves in ring galaxies may take about 3 x 10 8 years to propagate through the disk, and simulations suggest two or three such waves may do so before the phenomenon disperses (section 3.1). Collisionally induced bars and spirals may be longer lived (section 5.6). Thus, waves may make a non-negligible contribution to disk SF, despite the fact that they are much less spectacular than mergers.
Although star-forming tidal features in systems like the "Antennae" or "Super-antennae" have received a good deal of attention (section 5.4), significant amounts of SF are probably rare in tidal features, since gas is more generally dispersed rather than concentrated in such structures (see e.g., Schombert et al. 1990). Several sections above discussed systems with HI tidal features with few or no corresponding stellar features, and no on-going SF. It is unlikely that tidal features contribute much to the net SF in the universe. However, there are a couple of situations where tidal SF is more common than usual. The first is in mergers between equal, gas-rich progenitors (see section 6). The Antennae, and other objects on Toomre's classic list of merger candidates (1977) are examples (Schweizer 1978, also Duc and Mirabel 1997, Duc et al. 1998, Deeg et al. 1998).
The second category, Hickson compact groups, has been pointed out more recently by Hunsberger, Charlton and Zaritsky (1996 and section 5.4). The Hickson groups have already been mentioned several times above. Hickson (1982, 1993, and Hickson et al. 1992) assembled his list of groups using Palomar survey prints. Membership was determined on the basis of morphological criteria, especially compactness and isolation. It has since been demonstrated that most of the groups are probably gravitationally bound. These groups typically have a few to half a dozen members, and given their compactness, the merging time must be relatively short. It was for this reason, and the fact that many of the groups contain obvious interactions, that Hunsberger et al. selected them for a study of SF in tidal structures.
Using R-band images of 42 groups, Hunsberger et al. find 7 containing tidal arms or tails. Within these tidal features they discovered 47 knots or stellar concentrations, which they identify as newly formed dwarf galaxies, as noted in section 5.4. This identification is based on the luminosity of the knots, from which they estimate the knot mass. Hunsberger et al. interpret their finding of an average of 3 dwarf galaxies per tidal arm as confirmation of the models of Elmegreen et al. (1993). Followup studies, e.g., imagery in optical and infrared bands, are needed to determine the effects of reddening and obtain age estimates. If the ages of the stars in the knots matches the kinematic age of the tidal arms, we would have strong confirmation of the tidal dwarf formation theory.
The derivation of these kinematic timescales is independent of stellar evolutionary timescales, so the comparison between the two is potentially very interesting. In principal, this comparison between dynamical models, and stellar population models for local SF regions, could allow us to reconstruct the circumstances and the environment in which the SF occurred. Such studies would, provide powerful clues as to why specific forms of SF (e.g., formation of clusters, dwarf galaxies, etc.) occur at particular times and locations.
We have considered several examples of this comparison in preceding sections, including the following. 1) The symmetric ring galaxies are among the simplest (section 3.1). Theory and models predict that in these objects the collision has generated an outward propagating wave in the primary disk, and the wave compression triggers SF. Luminous young clusters in the Cartwheel and other rings, and strong color gradients behind the ring, indicative of an aging starburst population, strongly support these ideas. 2) Dynamical models also show the buildup of large clouds in tidal tails (e.g., Elmegreen, Kaufman and Thomasson 1993), and help us understand how dwarf galaxies could be produced. 3) Models of induced bars and resonance rings help us understand nuclear starbursts, though as yet with less predictive power in terms of the probable SF history (see section 5.6). These and other examples also show that the more detailed and unique the tidal structure in a given system, the more specific the modeling (both types) can be, and the "dating" will be correspondingly more precise.
There has been and should continue to be much progress in this area, especially with a wealth of new Hubble Space Telescope observations, infrared, radio continuum and 21 cm. line observations. The high resolution of HST, in particular, has allowed the study of individual star clusters in a number of colliding, merger remnant and starburst galaxies (see e.g., Miller et al. 1997, Whitmore et al. 1997, Kundu and Whitmore 1998, Schweizer 1998, and the recent text of Ashman and Zepf 1998). In one spectacular case, the "Antennae" galaxies, over a thousand individual clusters have been identified (Whitmore & Schweizer 1995, Schweizer 1998). With sufficient color data it is possible to estimate cluster ages, and learn about the history of starbursts in such sytems. Thus, this work is beginning to teach us much about the modes and mechanisms of induced SF. One impediment to progress is the fact that, because multi-wavelength observations are required, the work is data intensive and time-consuming. Moreover, many individual systems must be studied to derive reliable generalizations.
Mid-infrared observations provided by the Infrared Space Observatory (ISO) reveal dust heated by SF (or nuclear) activity, and are another source of exciting new information on SF morphologies, especially those buried in dust (see e.g., Genzel et al. 1998). a recent ISO study found that "the most intense starburst in this colliding system .... (is) in an off-nuclear region that is inconspicuous at optical wavelengths (Mirabel et al. 1998). This result, and similar results on extinction in merger remnants, provide important cautionary notes on the dangers of over-generalizing results based on optical observations. These cautions are especially important for high-redshift studies.
7.4. Mechanisms and Modes
7.4.1. Star Formation Enhancements
Galaxy collisions drive starbursts and starburst waves, but what can studies of collisional systems teach us about the general mechanisms of SF? Before delving into these questions let me summarize current thinking about large-scale SF in isolated disk galaxies. It is widely believed that in isolated disk galaxies the sizes and masses of star-forming molecular clouds are determined by the scale of the gravitational instability in the gas disk (e.g., the reviews of Kennicutt 1990, Keel 1991, Elmegreen 1992). According to this theory, if the gas exceeds a threshold density (or surface density) in the disk, then gravity can overcome pressure and shear in some range of wavelengths (BT, section 6.2). This leads to the formation of bound clouds on the scale of the most unstable wavelength. Kennicutt (1989) has presented evidence supporting the existence of the surface density threshold, and a number of recent studies support this picture (e.g., Caldwell et al. 1994, Skillman 1996, 1997, Kenney and Jogee 1997, Kennicutt 1998b and references therein), though some recent studies of nearby galaxies also suggest differences in detail between observations and the theory. Above threshold, the SFR appears to scale as a power-law function of the gas surface density with an exponent of about 1-2. This is the Schmidt Law (Schmidt 1959), which can be derived from cloud agglomeration models together with the assumption that SFR is proportional to cloud mass, for example. Kennicutt and others have argued that unperturbed galaxy disks are self-regulating, so that the gas surface density is maintained near the threshold value at all radii (also see e.g., Struck-Marcell 1991, Dopita 1990, Dopita & Ryder 1994).
In a gas disk the local threshold surface density is given by the expression (e.g., Kennicutt 1990, BT sec. 6.2),
where a is a constant of order unity, c is the velocity dispersion of the gas clouds, and k is the radial epicyclic frequency (see sec. 3.2). The gas surface density declines with radius as S µ 1/r in most gas disks (e.g., Struck-Marcell 1991), as does k when the rotation curve is flat (i.e., when the rotation velocity is constant). Moreover, c does not vary greatly with radius. Thus, based on these observations, it appears that S/Sc , the ratio of surface density to the local critical value does not vary greatly with radius.
However, there is no a priori reason to believe that self-regulation or the Schmidt Law obtain in galaxies that are driven far from equilibrium by collisions. In many cases the induced SF seems too nonlinear to be described by a Schmidt Law, and appears to depend on additional variables, like the kinematic disturbance (Keel 1993). A closely related idea is that of Elmegreen et al. (1993, Elmegreen 1994b) who suggest that the local turbulent velocity dispersion plays an important role in setting the mass scale of the star-forming clouds. The larger mass clouds, formed in more turbulent conditions, are more resistant to the disruptive effects of young stars, and so, it is argued, can sustain a higher SF efficiency. High-resolution observations of the atomic and molecular gas distributions in many systems are needed to determine where the Schmidt Law is applicable, and what is the role of other parameters.
In collisions where the disk is highly disrupted, the role of the gravitational instability as the primary organizing force for induced SF can also be questioned. Yet, while suppression of SF is the likely short-term outcome in such extreme cases, the presently available evidence suggests that gravitational instability probably has a major role wherever SF is enhanced. For example, gravitational instabilities can build larger cloud complexes in large-scale spiral density waves (see the review of Elmegreen 1992). Elmegreen, Kaufman & Thomasson (1993) have further suggested that gravitational instabilities in collisionally driven waves could build superclouds, with masses comparable to dwarf galaxies, which in turn, produce more stars (also see Barnes & Hernquist 1992 and the discussion of the previous section). Gravitational instability theory has also been extended to describe strong central starbursts in ultraluminous infrared galaxies (ULIRGs, see Elmegreen 1994b), and starburst rings at inner Lindblad resonances (Elmegreen 1994a). Thus, with extensions, the process may play a role in essentially all types of interaction or merger induced star formation.
However, other processes have also been suggested as the primary drivers of SF in more extreme cases of disequilibrium. These "violent" modes of SF include the direct triggering by cloud collisions or large-scale shocks. Perturbation analyses and cloud collision simulations in the 1970s inspired the idea that strong shocks can compress clouds and directly trigger (small-scale) gravitational collapse and SF. This idea has been studied in a number of simulational studies of galaxy collisions (e.g., Olson & Kwan 1990a,b, Noguchi 1990, Mihos et al. 1993, Mihos and Hernquist 1994, 1996). However, observations of collisional systems suggest that in some cases this process does not play an important role in stimulating SF. The symmetric ring waves provide one class of examples. The vigorous SF observed within the outer ring of the Cartwheel galaxy is apparently not caused by this mechanism, since the velocity discontinuity across the ring is modest (Higdon 1993, 1996). Moreover, in several ring galaxies with gas bridges connecting them to their collision partners, there is little or no young star population in the bridge (Appleton et al. 1996, Higdon 1996, Smith, Struck, and Pogge 1997). Yet, these bridges are most likely produced by direct, high velocity collisions, which would produce strong shocks.
Jog & Das (1992, also Jog & Solomon 1992) have suggested another "violent" mechanism, that is, SF triggered by cloud crushing in regions where the intercloud medium reaches unusually high pressures . In this picture, even as massive clouds are broken down, continuing SF could be driven in the cloud remnants or inflowing clouds, increasing the net SF. This process is probably most important in the centers of galaxies with strong radial gas inflows, and where young star activity and photoheating from AGN are additional sources of pressurization. Specifically, this idea was proposed to explain the high SF efficiency in ULIRGs, but may be relevant in other starburst environments as well. We expect transient high pressures in vigorous SF regions in collisionally driven disk waves, but the pressure is relieved after a short time by blowing out of the disk.
Pressurization and nonlinear enhancements from propagating SF seem to be "regulated" to modest amplitudes in unperturbed galaxy disks, but the existence of galactic winds in starburst galaxies (e.g., Heckman et al. 1993) provides one example of how they are not so tightly regulated in other environs. A recent symposium was held on the topic of "violent star formation" and its feedback effects in galaxies (Tenorio-Tagle 1994, also see the review of Shore and Ferrini 1995).
Besides their direct, disruptive effects, heating and other feedbacks may drive transient nonlinear dynamical instabilities in collisional galaxies. Scalo and Struck-Marcell (e.g., Scalo & Struck-Marcell 1986, Struck-Marcell & Scalo 1987) investigated cloud system dynamics with a model which included a cloud mass threshold for SF and young star feedbacks. The SF term in this "cloud fluid" model behaves as expected from the Schmidt law in its more quiescent mode. However, it undergoes a transition to burst-like behavior, which is more nonlinear than the Schmidt Law, above a critical density. This raises the possibility of a second density threshold above which the nature of SF is intrinsically time-dependent. The time between bursts and the burst duration in these models depend on intrinsic variables, independent of the gas consumption time (though subject to the effects of consumption).
Observationally, it is hard to distinguish between vigorous (Schmidt law) SF driven by high gas densities (or pressures), and a qualitatively different burst mode. What is needed is some measure of how nonlinear is the dependence of SF on gas density. Different observational techniques are used to find the atomic and molecular components of the cool gas. Each method has it's own ambiguities, and both generally have less resolution than we would like. Moreover, any region with strong SF does not become readily identifiable in the optical and near infrared until it has begun to disperse the gas, making it much more difficult to assess the conditions at the onset of SF. Nonetheless, the study of young post-starburst environments, will provide useful insights. For example, in the region interior to the star-forming ring in M82, it appears that not all the gas has been blown out, but that the clouds have been broken down to very small sizes (e.g., Lord et al. 1996).
7.4.2. Star Formation Suppression
Dynamical heating in galaxy collisions, and large-scale accretion processes can delay or inhibit SF in some regions via several mechanisms, though much less attention has been given to this than to enhancement mechanisms. For example, as already noted, strong bursts can trigger nuclear outflows and galactic winds (e.g., Heckman 1993, Lehnert & Heckman 1996). Outflows resulting from starbursts in the small companions in collisional systems may be particularly interesting, not only because they affect the SF history of the system, but also because, after a escaping relatively modest gravitational potentials, enriched gas may be thrown to relatively large distances. There it could contribute to the numerous absorption line systems observed in the spectra of quasars.
Within the galactic disks global compression waves are usually accompanied by comparable rarefaction regions. The region between the rings in ring galaxies provide striking examples (AS96). These waves can lower the gas density below relevant thresholds and terminate SF. Observations of the phenomenon can provide sensitive probes of threshold theories. To date, such observations have generally been consistent with the theory (e.g., in the Cartwheel, Lindsey-Shapley and Arp 10 ring galaxies, see Sec. 3.1).
Collisions of high velocity clouds with the gas disk can provide strong local shock heating and push cool gas out of the disk (Tenorio-Tagle 1981, Tenorio-Tagle et al. 1986, 1987). Both effects would tend to suppress cloud buildup and SF at least temporarily. Galaxy collisions involving two gas galaxy disks can push a great deal of gas out of the two galaxies. Most of this material will subsequently accrete back to the central regions of both of them, but at relatively high velocities. Struck et al. (1996) have suggested that the low level of SF within the inner ring of the Cartwheel galaxy may be partly the result of cloud disruption due to high velocity impacts of clouds infalling either radially from the disk or vertically from the bridge. Collisional galaxies are a unique environment for studying this process at a high flow rate, where global consequences are observable.
Models (Appleton et al. 1996, Struck et al. 1996, Struck 1997) suggest that the companion gas disk will be highly disrupted in near-central collisions between two gas disks, and subsequently reform via accretion out of the disk (also see Thakar & Ryden 1996). The heating involved in this process can delay the onset of SF while the gas mass builds up, and thus, setup a strong delayed starburst. While there are interesting hints in the available observations, the small companions in collisional systems have been relatively unexplored, and these predictions are largely untested. They potentially provide a valuable window on the process of galaxy disk formation.
In sum, galaxy collisions can be considered as Nature's experiments in how SF is affected by major rearrangements in the distribution and kinematics of the gas in disks. Spectacular enhancements and suppressions can result; the enhancements may be sufficient to account for a large fraction of the stars formed in the universe. The collisional disturbance can temporarily destroy the self-regulated state that controls SF in isolated galaxies, providing unique opportunities for studying individual SF processes when they are not regulated by couplings to other processes. The list of such processes is rather long, and their separate and synergistic roles are net yet well understood. There is much work yet to be done.