EXTREMELY LUMINOUS FAR-INFRARED GALAXIES - A RADIO PERSPECTIVE

ABSTRACT

We have imaged a sample of 27 extremely luminous far-infrared galaxies (ELFs - also known as ultraluminous FIR galaxies) using the Very Large Array in compact and extended configurations. These observations probe through the obscuring dust and reveal details of the nuclear activity. Knowledge of the nuclear structure and spectral indices allows us to test the relative importance of starbursts and active galactic nuclei (AGNs) for powering the prodigious far-infrared (FIR) luminosity of these objects. We have also used the Parkes-Tidbinbilla Interferometer to search for compact high brightness temperature radio cores which are characteristic of AGN activity. We find that ELFs frequently show clear signs of both starburst and AGN activity, in contrast to other recent results suggesting that almost all ELFs are powered by compact nuclear starbursts. ELF samples display a radio-FIR correlation whose small scatter is characteristic of normal spirals. Interestingly, we see a tendency for ELFs to lie slightly below the normal correlation, being more radio-quiet or FIRÒ-loud than normal spirals. We explore the consistency of starburst and other models that attempt to explain the ELF radio and FIR emissions, and point out some of their strengths and weaknesses.

Subject headings: galaxies: active - galaxies: Seyfert - galaxies: starburst - infrared: galaxies - radio continuum: galaxies

1. INTRODUCTION

Extremely luminous far-infrared galaxies (ELFs) are one of the most exciting and unexpected discoveries to come from the IRAS sky survey. These optically unremarkable galaxies distinguish themselves from other galaxies and active galactic nuclei (AGNs) by generating prodigious bolometric luminosities (1011.0 - 12.5 L) that are radiated almost entirely in the far-infrared (FIR). After their initial discovery by a number of groups (e.g., the review by Norris, Allen & Roche 1988) they have been studied intensively and now much is known about them. Spectacular hyperluminous examples (with LFIR ~ 1013 L) are now being discovered, making these among the most luminous known objects. Efficient survey strategies are being developed, and so we can expect many more ELFs to be found in follow-up work to the IRAS survey in the near future.

Our understanding of ELFs comes largely from works that address questions like: are they just one type of object? What are their spectral types, space densities, dust masses, gas kinematics and distributions? What leads them to radiate such large luminosities? Interactions frequently seem to be important - what events lead from interaction to large LFIR? Others seem not to be interacting - what else serves to trigger such activity? And a particularly important question is: how do ELFs relate to the many classes of AGNs?

The answers comprise the framework in which we will interpret our work. Optical spectra reveal two types of object that produce extreme FIR luminosities (e.g., Norris et al. 1988a). The majority have spectra classed as "starburst", dominated by powerful low-excitation narrow emission lines from H II regions photoionized by many O and B stars. The remainder are mostly classified as "Seyfert", having powerful high-excitation lines like [O III] and [NeÒšIII]. These are produced by clouds that may be photoionized by a central power-law optical-UV continuum source or by shock-excitation by plasma outflows. Although a minority at lower luminosities, the AGN fraction increases with luminosity to become the dominant spectral type for ELFs with LFIR > 1012 L (e.g., Sanders et al. 1988b). However, even in this range there remain some starburst-driven galaxies (Allen et al. 1991). Much recent debate has centred about the relative contributions made by these two different sources of luminosity to the radio and FIR emission.

The FIR-excess is caused by large quantities of dust that reprocess photons from optical and UV wavelengths to the FIR. Grains become warm upon absorbing optical or UV photons and then re-radiate a modified black body spectrum that peaks in the FIR. Evidence for this thermal FIR origin comes in many forms, such as the FIR spectral shape, the presence of emission from polycyclic aromatic hydrocarbons (PAH), absorption features in the near infrared from silicates, and a local minimum at around 1 m that rules out synchrotron emission (e.g., Telesco 1988; Bregman 1990). Further, the most luminous ELFs have Sy2 classification and Sy1s are conspicuously absent, which suggests that large quantities of dust may obscure our view of the AGN core, where the Sy1 broad lines are produced. This is consistent with the large optical depths (~ 4 to several magnitudes at V band, e.g., Norris et al. 1985; van den Broek et al. 1991) that are often found towards the nuclei of ELFs, using the near-IR colours and the Balmer decrement.

Unfortunately, this photon "laundering" hides the nature and origin of the original optical and UV photons. This has led to the debate over the relative contributions to the ELF phenomenon made by starbursts and AGN activity. For example, Colina & PÓ?rez-Olea (1992) and Sopp & Alexander (1991a) argue that ELFs are powered predominantly by starbursts. This followed from high-resolution (0.4" to 0.14") radio images of five ELFs which revealed that the radio emission is mostly from a circumnuclear starburst, not a compact AGN core. Norris et al. (1990) argue for a mix of starbursts and AGNs in a larger sample of ELFs, since 0.1" radio observations sometimes revealed the presence of high brightness temperature AGN-like cores. The ELFs with AGN cores also showed AGN-like optical spectra and those that lacked cores showed starburst spectra. Therefore, both starbursts and AGNs are present in ELFs. In contrast, Sanders et al. (1988b) argue that AGNs dominate the most luminous ELFs, for a number of reasons. AGN-like spectra are common, the NIR colours span a wide range, including those typical of AGNs, the FIR luminosities are similar to quasars, and the large ratio of LFIR / M(H2) requires an efficient source of luminosity. Alternatively, Harwit et al. (1987) argue that the conversion of kinetic to thermal energy in colliding galaxies can power ELFs, and claims that the starburst model of ELFs is flawed as it predicts a correlation between LFIR and Loptical that is different from that observed. However, Sanders, Scoville & Soifer (1991) criticise the proposal of Harwit et al. on the grounds of time scales. They argue that two spirals would need to collide nearly face-on to convert the energy fast enough to produce the required large luminosity.

Despite the difficulties, the debate is important for a number of reasons. First, the most extreme ELFs are among the most luminous objects known (e.g., compare the ELF IRAS 00182-7112 of Heckman, Armus & Miley 1990 to the luminous quasar of Hagen et al. 1992) and there is an intrinsic appeal in knowing what drives this spectacular phenomenon. Second, chemical evolution of normal galaxies will be profoundly affected if all galaxies go through a starburst-ELF phase. Third, the possible relationships between ELFs and various other AGNs are constrained by similarities between their engines. Finally, galaxy dynamics are affected by the dramatic redistribution, that seems characteristic of ELFs, of the interstellar medium (ISM) into the centre. Despite having a space density that is 104 times less than non-ELF galaxies, the ELF phenomenon has the potential to affect the development of almost every common type of galaxy that we see around us.

The progenitor galaxies that become ELFs are gas-rich spirals (e.g., Sanders et al. 1988b). Much of the ISM is thought to be driven into the central kpc by interactions or by a bar instability, and provides there a large reservoir of relatively high density gas. Much of this gas is in molecular form and has been imaged with high spatial resolution by mm-arrays (e.g., Sanders et al. 1988a) revealing massive concentrations of gas centred on the optical nuclei. The 109-10 M of gas accounts for typically one third of the total molecular mass in these galaxies, and for almost all of the dynamical mass in the nuclei (Scoville et al. 1991). This material is available both for star formation and to fuel AGN activity. The efficiency of star formation (measured by the ratio LFIR / Mgas) is typically very high compared to normal galaxies (e.g., Chini, KrÓÌgel & Kreysa 1992).

Whilst interactions are clearly important (e.g., Sanders et al. 1988b), other trigger mechanisms may also be needed since not all ELFs seem to be interacting (e.g., Lawrence et al. 1989). However, the problem of providing such a trigger is hard, and the solution probably requires an understanding of the processes by which stars form. Such knowledge is only beginning to emerge.

The ELF phenomenon is clearly a short lived one as the gas supply is not sufficient to maintain the currently observed luminosities beyond ~ 1% of a Hubble time. Thus, these galaxies must appear very different for most of their lives. It is a major goal to understand what they were before becoming ELFs and what they will evolve into afterwards. There is wide agreement that the progenitors of ELFs were probably gas-rich normal spirals, mainly because of the large molecular gas content of ELFs. Their immediate future, according to Sanders et al. (1988b, c) is destined to be spent as normal quasars. According to this scheme, the collision of two normal spirals fuels a pre-existing quiescent black hole and triggers a powerful nuclear starburst. The starburst-driven winds blow away the enshrouding dust and lay bare the quasar core as the starburst dies away, at which time the object should become a normal quasar.

Radio observations have the ability to probe through the present dense obscuring clouds of dust and gas to reveal details of the nuclear activity and so help constrain some of these suggestions. These observations can reveal the presence of starburst and AGN activity from the radio morphology, spectral indices, and variability. Starbursts tend to be extended over many hundreds of pc, have steep spectral indices, and have brightness temperatures of ~ 100 to 104ÒšK. In contrast, AGN activity has distinctive compact radio cores with high brightness temperature, and shows rapid variability, and sometimes produces jets and lobes. Radio imaging has been used previously by Condon, Huang & Yin (1991) using the Very Large Array (VLA) in its extended (A) configuration which provided 0.25" resolution. They looked at the morphology, spectral indices and brightness temperatures of a uniform sample of ELFs drawn from the Bright Galaxy Sample (BGS). They concluded with the well-argued view that ELFs are almost all due to starburst activity, in interesting contrast to the QSO-progenitor scheme of Sanders et al.

We have selected a similar, though largely independent ELF sample, and made images using the VLA in both compact and extended configurations. We obtained improved sensitivity to possible extended low surface brightness lobes characteristic of radio galaxies, and the high-resolution images enable us to look for morphological and spectral index indicators of AGNs and starbursts. These high-resolution observations have sufficient surface brightness sensitivity to detect radio emission from typical extended starburst regions, which range from tens of kelvin to 104 K (e.g., Hummel 1981; Condon et al. 1991; Forbes et al. 1994). We also used the long-baseline Parkes-Tidbinbilla Interferometer to look for compact high brightness temperature AGN cores. Along with optical spectral classifications these observations together paint a picture of ELFs that interestingly differs in detail from recent works by finding frequent evidence for AGN activity, as well as the usually dominant starbursts.

A reader wanting a quick overview should look at Figure 2 and read Ò?Ò? 4.2, 5.1 and 5.2.

The parameters H0 = 75 km s-1 Mpc-1 and q0 = 0.5 are adopted throughout this paper.

2. SAMPLE

As ELFs radiate the bulk of their luminosity at FIR wavelengths, finding surveys are most efficiently conducted using the IRAS 60 or 100 m bands. Our sample is therefore FIR-selected, drawn mainly from the IRAS galaxy sample of Allen et al. (1991). These authors selected a sample of 1178 IRAS galaxies that were detected at both 60 and 100 m, that had relatively cool 25 to 60 m colours (to reject stars), that were not identified with a Galactic or SMC object, and that lay within three small regions of sky at high Galactic latitude. They obtained redshifts and, where possible, optical spectral classifications for 475 of these. They chose those that were smaller than ~ 1' in the ESO sky survey, since their interest lay with the most extreme infrared galaxies.

We have now selected for further observation a subset of 25 of these galaxies and two ELFs from a related project. First, we chose the 10 most FIR-luminous galaxies from Field 3 that had 60 m flux densities greater than 1ÒšJy, and that lay within 22h40m < RA < 2h (B1950). All 10 are ELFs, having LFIR > 1011ÒšL. By including a further two galaxies (01526-2407 and 23245Ò-3634, ranked 23rd and 24th in FIR luminosity) the sample then included all the ELFs that were classified optically as starbursts as at 1989 January, and that satisfied the above constraints on the locations, FIR flux densities and luminosities. Next, we also included the two most spectacularly luminous ELFs (00188-0856 and 23046-3454) from a related project. As above, these have 60 m flux densities greater than 1ÒšJy, and lie within 22h40m < RA < 2h. Both galaxies have LFIR > 1012ÒšL. Finally, we included 13 ELFs from Field 1 of Allen et al., selected for having LFIR > 1011ÒšL and for being conveniently located for observing The resulting sample is a representative selection of extreme objects, mostly from Allen et al., selected for having large FIR luminosities and 60 m flux densities greater than 1ÒšJy.

This ELF sample is directly comparable to ELFs selected from the Bright Galaxy Sample by Condon et al. (1991) (BGS ELFs hereafter). The BGS ELFs were selected using a similar procedure, and they have similar FIR colours and luminosities and AGN fractions to our present sample. These samples seem to differ only in their FIR flux density limits, with the Allen et al. sample going about a factor of five to ten deeper, approaching the IRAS sensitivity limit at 60 and 100Òšm. Our sample therefore extends to larger distances than the BGS ELFs, and may exclude the nearer, optically large ELFs.

3. OBSERVATIONS

3.1. VLA Observations

The ELF sample was observed with the VLA at two frequencies (1.49 and 8.44ÒšGHz) in both compact and extended configurations. These configurations had maximum baseline lengths of 3.4Òškm and 22.8 km (C and A/B arrays). We measured the total radio emission and spectral indices, and looked for structures that may distinguish starburst from AGN emission. We used the low-resolution, C-array observations to measure the total radio emission and to reveal the presence of companions and extended, low surface brightness radio lobes if present. The high-resolution, A/B array, observations then provided a more detailed view of the core. They were made to look for the relative importance of core and extended emission, and to look for morphological indications of starburst versus AGN activity. The surface-brightness sensitivity and resolution of these A/B-array observations should allow starburst regions brighter than ~ 25 K and larger than ~ 200 to 2000 pc to be detected and resolved. They can also detect compact high brightness temperature AGN cores stronger than ~ 1 mJy. Distinctive structures such as the diffuse extended emission produced by starbursts or the core-jet morphology that often characterises AGN emission can provide the means to distinguish starburst and AGN processes.

The galaxies were observed between 1989 August and 1990 July. We made 20 min snapshots centred on each ELF, with nearby phase calibrators observed before or after each programme source. Observations with the C array provided a resolution of 11 arcsec at 1.49ÒšGHz and, with a largest fringe spacing of ~ 7 arcmin, should not have missed any extended emission. The resolution provided by the A/B-array was 0.5 arcsec at 8.44 GHz, corresponding to 800 pc at the median redshift of the sample. We used a bandwidth of 100ÒšMHz, with system temperatures of ~ 50 K at 1.49 GHz and 35 K at 8.44ÒšGHz. We recorded both hands of circular polarization from which we made images of total intensity.

Calibration and imaging were carried out using the AIPS software, using standard methods. The flux-density scale was based on that of Baars et al. (1977), and assumed that 3CÒš48 was 15.2 Jy at 1.49 GHz and 3.14 Jy at 8.44 GHz, and that 3C 286 was 14.4 Jy at 1.49ÒšGHz and 5.19 Jy at 8.44 GHz. The data were imaged using uniform weighting to obtain high resolution and to reduce sidelobe levels. The whole primary beam was imaged, and confusing sources and programme sources were cleaned. In the few cases where the ELF was initially undetected, we then used natural weighting to improve sensitivity, which mostly yielded detections. Self-calibration was used for all 8.44 GHz observations in the A/B configuration in which a source stronger than ~ 2 mJy was present. It was also used for the three 1.49ÒšGHz A/B-array observations in which the ELF initially showed structure, and was used for many 1.49 GHz CÒ-array observations. We have applied coherence corrections to the flux densities from any 8.44 GHz A/B-array observations that were not able to be self-calibrated. The corrections were estimated by comparing the calibrator observations before and after each programme source. This yielded a measure of the phase drift, from which the decorrelation could be estimated for each programme source. The phase instability was caused mostly by the atmosphere and so we adopted a standard atmospheric model, consisting of frozen-in Kolmogorov turbulence. We used Fig. 1 of Buscher (1988) to derive the coherence corrections for the 20 min time scale of the programme source observations from the decorrelation measured on 1 h time scales using the calibrator sources. The coherence corrections that we applied were 1.28 for 00148-3153, 1.11 for 00494-3056, 1.11 for 01077Ò-1707, 2.00 for 01199-2307, 1.28 for 01494-1845, 1.19 for 01526-2407, 1.49 for 13579-1848, 1.23 for 14207-2002, 1.25 for 22525-2624, and 1.25 for 23046-3454. The coherence loss was <~ 1.04 at 8.44 GHz for the C-array observations, and so we have not applied coherence corrections to any other measurements. A selection of typical images is shown in Figure 1, along with the two objects that displayed considerable structure.

We measured the peak and integrated flux densities of the ELFs both directly from the cleaned image, and by fitting a 2Ò-D Gaussian and a baseline. We subtracted the baselines from the directly integrated flux densities, after which the two methods typically agreed within 5%. RMS noise levels in the cleaned images were typically 0.28 mJy beam-1 at 1.49ÒšGHz and 0.07 mJy beam-1 at 8.44 GHz. These sensitivity limits correspond to 5- brightness-temperature limits from 0.9 K with C array at 1.49 GHz to 25 K with the A/B array at 8.44ÒšGHz. The resulting flux densities are estimated to have uncertainties of 12% random multiplicative (or 26% if a coherence correction was applied), with 0.3 mJy random additive, and a systematic multiplicative uncertainty of 11% due mainly to uncertainty in the Baars et al. (1977) flux density scale.

The flux densities measured for 12302-2020 do not agree within these uncertainties. No fault with the data reduction could be found, and so this source probably varied during the interval between the C-array and the A/B-array observations.

The results are presented in Table 1 which, for convenience, includes data on a further 11 similar galaxies that were observed as part of another project and were not included in our present study.

3.2. PTI Observations

Many of our ELFs (21), along with 19 BGS ELFs, were also observed using the Parkes-Tidbinbilla Interferometer (PTI), which is described in detail by Norris et al. (1988b). The interferometer consists of the 64 m antenna at Parkes with the 70 m antenna at Tidbinbilla over a 275 km baseline. It provided 0.10" fringe spacing at 2290 MHz, or 0.14" at 1665ÒšMHz. The galaxies were typically observed for 20 min each with two 5 MHz bands sensing one hand of circular polarization. This yielded a 5- sensitivity limit of ~ 3 mJy, or a brightness temperature limit of 105 K. The observations are described in detail by Roy et al. (1994). The PTI is an acute filter which is sensitive to compact high-brightness temperature AGN cores but is blind to extended 104 K emission that is typically produced by circumnuclear starbursts. Thus a detection indicates unambiguously that an AGN is present. The results are presented in Tables 1 and 2.

3.3. Other Data

We used spectral classifications made by Allen et al. (1991) using line-ratio and line-width diagnostics, and FIR data were taken from the IRAS Point Source Catalog (Version 2). These data are shown in Tables 1 and 2, along with other useful information.

4. RESULTS

4.1. Starburst vs. AGN: The Radio-FIR Correlation

A powerful diagnostic for distinguishing starburst from AGN activity is provided by the common and extraordinarily tight radio-FIR correlation that is displayed by surprisingly diverse galaxy types. The correlation is often thought to be caused by star formation activity. In starburst galaxies, FIR is produced by dust heated by young massive stars, and the radio emission comes indirectly from the supernovae into which these stars evolve. AGNs, on the other hand, often depart from the correlation, tending to be more radio-loud than starburst galaxies for a given FIR luminosity (e.g., Wilson 1988; Sopp & Alexander 1991b; Roy et al., in preparation). ELFs have been shown previously to display a tight radio-FIR correlation (e.g., Condon et al. 1991), and this probably indicates that star formation activity makes a major contribution to the radio and FIR emissions. We confirm that this tight correlation is displayed by our present sample, particularly at the higher frequency (Figure 2). This suggests that AGN activity is not usually the dominant source of power for ELFs.

Suprisingly, we find from the present C-array survey that ELFs actually depart slightly from the normal radio-FIR correlation at 8.44 GHz. The ELFs tend to be displaced from the normal correlation in the sense of being radio-quiet, or FIR-loud. This effect is also visible in the systematically selected ELF sample of Condon et al. (1991), and the evidence for these offsets can be seen in Figure 2.

We have chosen from the literature three control samples of normal galaxies, selected in various ways, against which to compare the distribution of FIR / radio ratios of both our present ELF sample and the BGS ELFs of Condon et al. (1991). Our ELFs display a median FIR / radio ratio, q, of 3.03 at 8.44 GHz and a dispersion, , of 0.19 in the log (excluding the Seyferts). For comparison, they are a little more radio-quiet than the control samples, which display FIR / radio ratios, q, of 2.87, 2.84, and 2.76 for the BGS galaxy sample, the sample of Wunderlich, Klein & Wielebinski (1987) and the sample of Marx et al. (1994). These control samples have 293, 105 and 26 objects respectively. The dispersion displayed by the ELFs is similar to that of the control samples, which display dispersions of 0.19, 0.19, and ~ 0.20 respectively. For these comparisons, we discarded galaxies from the control samples that were classified as Seyfert as this class of object is known to depart from the normal correlation (e.g., Wilson 1988; Roy et al., in preparation).

We now present the tests that quantify the difference between the ELFs and normal galaxies, revealing that it is formally significant. To ensure independence between the control and ELF samples we exclude all ELFs (galaxies with LFIR > 1011 L) from all three control samples for the purposes of the following statistical tests. This is justified since we want to compare normal spirals (those with LFIR < 1011 L) to ELFs (those with LFIR > 1011ÒšL). Further, we have excluded from the control and ELF samples those ~ 10% of galaxies that are classified as Seyfert in the NASA Extragalactic Database as at 1994 AprilÒš26, or that have a PTI-detected compact radio core, since AGN activity is already known not to display the normal radio-FIR correlation (e.g., Wilson 1988; Roy et al., in preparation).

We used survival analysis two-sample tests to compare the distributions of the FIR / radio ratio, and found a very significant difference between the ELFs and all three control samples. Comparing our ELFs to the BGS galaxies, the weakest test returned P = 2.5% (Logrank test). Comparing our ELFs to the Wunderlich et al. sample, the weakest test result was P = 1.3% (Logrank test, after conservatively dropping four galaxies that had upper limits on q and keeping the 14 with lower limits, since survival analysis can handle one type of limit only). Comparing our ELFs to the Marx et al. (1994) spiral sample, the weakest test result was PÒš= 0.05% (Gehan permutation variance test). Thus, the difference between the ELFs and the control samples is formally significant for all three comparison samples.

The BGS ELFs, like our present ELF sample, also lay significantly below the normal correlation. The weakest tests returned P = 2.7% comparing the BGS ELFs to the Wunderlich et al. sample, 1.2% against the Marx et al. 1994 sample, and 7.6% against the BGS galaxies.

This tendency for ELFs to be radio-quiet at 8.44 GHz has not been widely recognised, although it is visible at low significance in the data of Condon et al. (1991). It may present an insight into ELFs and we explore some possible explanations in Ò? 5.1.

Three other ELFs (01569-2939, 13305-1739 and 14254-2655) lie far above the correlation. All three are classified as Seyferts, and one also harbours a compact radio core. In these cases the radio emission, at least, is most likely to be dominated by nonthermal AGN emission.

4.2. Compact Radio Cores in ELFs

Many ELFs in both our sample and that of Condon et al. (1991) were observed with the PTI, and flux densities or limits are given in Tables 1 & 2. Galaxies were selected for observing on various grounds (e.g., presence in the FIR-selected Seyfert sample of de Grijp, Miley & Lub (1987), in the Markarian survey, etc) and the sample observed is therefore somewhat heterogeneous. However, approximately 18% (7/40) of these ELFs were detected by the PTI, showing that ELFs frequently harbour AGNs and that our view of the cores in these galaxies is relatively unobscured at GHz frequencies. This confirms the suggestion by Colina & PÓ?rez-Olea (1992) that such cores might be found in the BGS ELFs. It is also mostly consistent with the VLBI survey by Lonsdale, Smith & Lonsdale (1993) which found that compact radio cores are common among the BGS ELFs. It is in contrast to the conclusion of Condon et al. (1991) who found convincing evidence for an AGN in only one case (MknÒš231).

It is unlikely that the PTI detections could be due to radio supernovae. A supernova would need a radio luminosity of ~ 104 times Cas A to be detected by the PTI at the smallest redshift of this sample. Such a powerful explosion is rare - even the hyperluminous object SN1979c (Colina & PÓ?rez-Olea 1992) would not be detectable by the PTI if placed at the smallest redshift of our sample. Further, the survey of starburst galaxies by Norris et al. (1990) using the PTI detected compact radio sources in only two out of the 57 starburst galaxies surveyed. At the same time, these authors obtained a much higher detection rate (21/84) for Seyfert galaxies, which presumably contain AGNs. Their galaxies spanned a similar redshift range to the present ELF samples, and so we expect that our present PTI detections most probably indicate the presence of AGN cores.

4.3. Spectroscopic Classifications

The spectroscopic classifications of Allen et al. (1991) reveal that ELFs characteristically have starburst, Seyfert, or LINER spectra. We list classifications in Table 1 for our present ELF sample, and in Table 2 for the BGS ELFs. The frequent occurrence of AGN-type spectra is further evidence that ELFs sometimes harbour AGN cores in addition to, or perhaps in place of starburst activity.

Further, we see a tendency for AGNs to be more common in galaxies of higher FIR luminosity. Taking all 388 galaxies classified by Allen et al. (1991), we show (Figure 3) that the fraction of AGNs increases with FIR luminosity. The fraction increases from ~ 9% for less luminous galaxies up to 60% for the several most luminous ELFs that have LFIR > 1012ÒšL. This is consistent with previous findings by, for example, Sanders et al. (1988b). In addition to the optical emission, in some cases a FIR component has also been identified with the AGN (e.g., Telesco 1988; Rowan-Robinson & Crawford 1989; Wynn-Williams & Becklin 1993).

Interestingly, even in the most luminous ELFs we are often able to obtain a relatively unimpeded view into the central few hundred pc of the nucleus where the narrow emission lines originate. In one case (01569-2939) we can even see right down to the broad-line region, located much closer again to the central continuum source (within ~ 20 light-days). This suggests that in many ELFs the dust obscuration that is responsible for the large FIR excess must be patchy. It is also inconsistent with the suggestion, by Barthel (1989) and others, that ELFs may be predominantly dust-obscured quasars.

4.4. Radio Structures

Most of the radio emission from our ELF sample originates typically within the central kpc, and there is in general no well resolved structure in our VLA images above the brightness temperature limit of ~ 1 K. These observations rule out the existence of radio jets or extended radio-galaxy-like radio lobes in our ELF sample. Such structures typically have brightness temperatures of 101-3 K (e.g., Fernini et al. 1993) and scale sizes (~ 1') that are well sampled by our C-array observations, and so they should easily be detected. It is very unlikely (P = 0.1%) that all 65 ELFs in our sample and the BGS ELFs would turn out to be radio-quiet if selected at random from a parent population in which 10% are radio-loud. (A radio-loud fraction of 10% was adopted to match to behaviour of the the UV-excess quasars studied by Peacock, Miller & Longair 1986.) Thus, in the ELF-QSO evolutionary scheme of Sanders et al. (1988b), the radio jets and lobes probably form after galaxies have evolved beyond the ELF stage by expelling the dust and gas from the nucleus.

We see extended structure in only one ELF (01077-1707) in our sample, shown in FigureÒš1. The diffuse kpc-scale structure and the peak brightness temperature (30 K) are typical for starburst emission and uncommon for AGNs. The PTI did not detect a compact radio core, ruling out the presence of a core brighter than 105 K at 2.3 GHz. The remainder of the ELFs (26/27) do not display structure in images from the extended VLA configuration. The general lack of structure in our ELF sample is in contrast to the BGS ELFs of Condon et al. (1991), of which about three quarters show some extended diffuse emission with low brightness temperature characteristic of starburst activity. The surface brightness sensitivity of the Condon et al. observations was similar to that of our 8.44 GHz A/B-array observations (1- = 9 K compared to 5ÒšK). The more complex structure found by Condon et al. is therefore likely to be due to the relative proximity of their sample (48 to 327 Mpc with a median of 130 Mpc, compared to 81 to 680ÒšMpc with a median of 340 Mpc for the present sample). Their smaller beam size (0.25 x 0.25" FWHM compared to 0.8 x 0.3" FWHM) should also contribute to the difference. The choice by Allen et al. (1991) to reject galaxies that appear in the ESO and MCG catalogues from their sample prevented any nearby large ELFs from entering the present sample, which should make structure more difficult to resolve in the present sample.

Despite the lack of structure, the 8.44 GHz A/B-array observations revealed that many sources were slightly extended. The amount of extension ranged from less than one tenth of a beamwidth, to about 1.5 beamwidths, and the median extension corresponds to a radius of ~Òš500Òšpc. Some amount of this extension may be due to beam-broadening due to atmospheric phase fluctuations, and so the actual source sizes may be smaller than those indicated here. We use the observed sizes later to infer magnetic field strengths in the central regions from the brightness temperatures of the synchrotron emission.

4.5. Core Spectral Indices

We found, surprisingly, that the spectral index of the radio core depends on the FIR / radio ratio, in the sense that the more radio-quiet ELFs tended to have flatter spectra (Figure 4). This correlation is significant at the 0.9% level of significance, using the Generalized Kendall's Tau test, and so is very significant. In contrast, if ELFs become radio loud due to emission from flat-spectrum AGN cores then one would expect the opposite trend, specifically that the more radio-loud ELFs should have flatter spectra. The observed trend is explained by the model proposed by Condon et al. (1991), in which the nuclear emission from ELFs is produced by a compact optically thick nuclear starburst. According to that model, in cases where the nuclear starburst is optically thick, the spectrum becomes flatter and the galaxy becomes more radio-quiet due to self absorption.

5. DISCUSSION

5.1. Why Do ELFs Depart from the Normal Radio-FIR Correlation?

We showed in Ò? 4.1 that, surprisingly, ELFs tend to lie below the normal radio-FIR correlation, being slightly more radio-quiet, or FIR-loud, than normal spirals. A similar effect was seen at lower frequencies by Condon et al. (1991) and by Staveley-Smith et al. (1992), and a number of explanations have been proposed.

5.1.1. The "Young Starburst" Model

Staveley-Smith et al. attribute the offset to the recency of the starburst. According to this model, the FIR emission is expected to be produced much earlier after the burst begins than the radio emission, which must await the explosions of the first supernovae. Young ELFs should therefore be more radio-quiet than normal.

However, this scenario seems unlikely on the basis of time scales. According to this model, the FIR-loud phase is expected to last ~ 107 yr until the radio emission catches up. For comparison, the typical lifetime of a starburst may be up to ~ 10 times longer, at 108 yr (e.g., Sanders et al. 1988a, and Ò? 5.3 below) and so only about one in every 10 ELFs should be at the FIR-loud stage. Such a small effect would not be sufficient to produce the observed offset in our present sample.

5.1.2. The "Superwind" Solution

This time scale problem of the young starburst model could be avoided if the ELF phase is made to last for much less than the total 108 yr lifetime of the starburst. Perhaps, by the time the first supernovae explode to produce the radio emission, the character of the galaxy may have changed dramatically. For example, starburst-driven winds may by then have expelled the obscuring dust and so caused the galaxies no longer to be classified as ELFs. This would increase the fraction of ELFs that are FIR-loud. We can again rule out such a scenario because of time scales. Wind speeds of ~ 50 000 km s-1 are needed to clear dust from the central kpc in the required 107 yr time scale, whereas much smaller top speeds (~ 1000 kmÒšsÒ-1, e.g., Heckman, Armus & Miley 1990) are actually seen.

Alternatively, dust can be destroyed by many common processes, such as by thermal sputtering, by being incorporated into stars, or by supernova shocks (e.g., Seab 1987). Sputtering needs temperatures much larger than those in molecular clouds in ELFs and so is probably unimportant. The formation of stars consumes only a small part of their parent clouds as most of the gas is blown off when protostars switch on, and so star formation is unlikely to destroy much dust. Supernova shocks, on the other hand, are estimated to destroy large amounts of dust on relatively short time scales; ~ 2x108 yr in the Milky Way. The rate is perhaps 100 times faster in ELFs given their much larger expected supernova rates. However, the large destruction rates are offset by the large production rates from the enriched ejecta of supernovae in ELFs. Unfortunately, the rates are highly uncertain and the best models predict steady-state dust concentrations that are inconsistent with the large amounts of dust seen in the Milky Way. Therefore, we cannot reliably estimate the effect of supernovae on the dust content of ELFs.

We might obtain an unobstructed view into ELFs if dust grains were to grow by accretion in molecular clouds to sizes that are too large to cause much absorption. Indeed, dust clumps have been recovered from cometary nuclei, but the process by which they form is not understood and we cannot estimate the effect on the dust content of galaxies. Thus, it is unclear whether dust could be destroyed in ELFs in the required 107Òšyr time scale.

5.1.3. The Optically Thick Starburst Solution.

Condon et al. (1991) suggest that ELFs are powered entirely by starburst activity, and that the nuclear starburst has a sufficiently large plasma density to become optically thick at low GHz-frequencies due to free-free absorption, making the ELFs appear radio-quiet. They further constrain their model to match the observation that ELFs become optically thin by 8.44 GHz. A similar model is invoked by Sopp & Alexander (1989) to account for the unusual low-frequency turnover in the radio spectrum of Arp 220. This model has much evidence to support it, but also encounters some problems. The advantages, argued by Condon et al., are that (i) the radio-quiet offset is naturally explained, (ii) the tighter correlation at the higher frequency is explained, (iii) the relatively flat radio spectral indices have a natural explanation in the free-free absorption, and indeed the flatter the spectral index, the more radio-quiet is the ELF, consistent with free-free absorption, and (iv) the brightness temperatures are observed to approach but never exceed 104 K, a value that is characteristic of starburst emission. Furthermore, in those cases where the brightness temperature approaches the maximum, the objects also tend to be more radio-quiet than normal. This is as expected, since to achieve the highest brightness temperature requires the objects to be optically thick.

However, the compact nuclear starburst model runs into three important difficulties. First, we observe optical AGN spectra in many ELFs, and so one cannot immediately discount the AGN from making significant contribution to the radio and FIR emission. Second, in many ELFs we detect compact high brightness temperature radio cores, and these are likely to be associated with AGN activity. These cores probably lie within the central few hundred pc (Roy et al. 1994), and their detection argues that any surrounding nuclear starburst is not very optically thick at 2.3 GHz. Third, one expects a huge H luminosity from the multitude of nuclear H II regions that must comprise a starburst of the proposed luminosity. (For the typical nuclear starburst conditions of 104 K, Ne = 108 m-3, radius of starburst = 500 pc, assuming a unity filling factor, we predict an H luminosity of 1.7x1037 W (4.4x1010 L).) Such a large H luminosity is seen in only the most luminous ELF known (IRAS F10214+4724), and other ELFs fall short of this value by more than an order of magnitude (e.g., Kennicutt & Kent 1983; Devereux & Young 1990; van den Broek 1992). One may invoke an order of magnitude obscuration by dust to bring the H flux within the observed range, but one must somehow also reconcile this with our ability to see narrow and broad lines in some ELFs. These lines originate within the central few hundred pc and should also be subject to similar obscuration.

A strong observational test has been carried out by Prestwich, Joseph & Wright (1994), who observed the near-IR line of Br from eight ELFs. Dust obscuration is an order of magnitude less at Br than in the optical, and they found the recombination line fluxes to be consistent with the compact nuclear starburst model in six out of those eight ELFs. From these, it seems that the H emission is indeed heavily absorbed. In the other two ELFs (ArpÒš220 and NGC 6240) an AGN component might be required. The presence of narrow and broad emission lines in some ELFs implies that the obscuring dust must be patchy.

5.1.4. The Initial Mass Function Slope Solution

Colina & PÓ?rez-Olea (1992) suggest that the slope of the initial mass function (IMF) differs between ELFs and normal galaxies. This alters the ratio of stars with mass near 5ÒšM, which produce the bulk of the supernovae, to the higher-mass stars, which produce the bulk of the FIR emission. Likewise, the relative numbers of low- and high-mass stars depend on the IMF mass cut-offs, and increasing either cut-off could produce the observed offset. Later, we derive magnetic field strengths that are up to 100 times larger than those in normal disc galaxies. Perhaps the large field strengths might inhibit low-mass star formation sufficiently to bias the IMF as required.

5.1.5. The Thermal Emission Solution

The FIR / radio ratio of Galactic H II regions is greater than that of normal spirals (as detailed below). The excess FIR emission from ELFs could therefore be explained if H II regions provide a greater fraction of the total radio and FIR output from ELFs than they do from normal galaxies. The typical FIR / radio ratio, q, of Galactic compact H II regions is 3.21 ÒÁ 0.08 at 8.44 GHz (using radio data from Wink, Altenhoff & Mezger 1982; see also Haslam & Osborne 1987). Thus, H II regions are more radio-quiet than normal spirals (for which q8.44ÒšGHz = 2.84 ÒÁ 0.05) and are also more radio quiet than the ELFs (q = 3.03 ÒÁ 0.04). One could therefore reproduce the observed ratio displayed by ELFs if they had a thermal fraction of ~Òš40%, assuming spirals are entirely nonthermal following work by Gioia, Gregorini & Klein (1982).

5.1.6. The Dust Opacity Solution

In ELFs, a large fraction of the optical output of stars is reprocessed down to the far-infrared by obscuring dust. This fraction is much larger than is seen in normal galaxies. The extra obscuring dust in ELFs is not expected to produce a corresponding increase in radio output and so might cause the observed FIR excess displayed by ELFs.

The size of this effect can be roughly quantified by comparing the FIR to the optical luminosities for normal galaxies and for ELFs. The ratio of LB / LFIR for the galaxy sample of Wunderlich et al. (1987) ranges from 0.2 to 44 with a median of 1.3, using B-band magnitudes from Paturel et al. (1989). In contrast, the ratio is much larger for the ten ELFs studied by Sanders et al. (1988b). For these extreme objects, the ratio is 10 to 102 with a median of 45. Therefore, normal galaxies radiate roughly equal luminosities at optical and FIR wavelengths, whereas in ELFs the bulk of the luminosity comes out in the FIR. ELFs should therefore be roughly a factor of two more FIR-loud for the radio emission than normal galaxies. This is similar to the observed factor of 1.6 found in Ò? 4.1, and so the extra dust opacity present in ELFs could cause the observed FIR excess.

5.1.7. A Biased Sample?

Our ELF sample is FIR-selected, and so at first sight this might select for objects that are FIR-loud. We argue that such a bias does not actually occur, by the following reasoning. Consider a parent population of ELFs that display a radio-FIR correlation, with a distribution of the FIR / radio ratio that is independent of the FIR flux density (a realistic situation). If one imposes a FIR flux-density limit then the ELFs that lie above the limit should display the same distribution of FIR / radio ratio as those below the limit. Provided that all ELFs in the FIR-selected sample are subsequently detected by the radio survey then the resulting distribution of the FIR / radio ratio will be representative of the parent population. There will not be a bias towards FIR-loud objects.

To verify this reasoning, we performed a Monte Carlo simulation that calculated the mean FIR / radio ratio of an artificial ELF sample before and after applying a FIR flux-density and luminosity limits. We constructed 1000 parent samples each of 1000 ELFs, with redshifts and luminosities distributed in a manner similar to real ELF samples. The artificial ELFs were randomly distributed in space with uniform space density out to a redshift of 0.5. They spanned FIR luminosities of ~ 1010 to 1012 L, with a uniform probability density in log(LFIR). This probability density is similar to that of our real ELF sample, and approximates the rapid increase of space density with decreasing luminosity that is typical of most luminosity functions. The FIR / radio ratio was chosen to be 2.84, to match that of the control sample of spirals from Wunderlich et al. (1987). Scatter was then introduced into the correlation by moving ELFs perpendicularly away from the normal correlation by an amount dictated by a Gaussian random variable with zero mean and a dispersion of 0.2 in the log. After calculating the mean FIR / radio ratio for each such parent sample, we applied a FIR flux-density limit of 1 Jy and a lower luminosity limit of 1011 L in the same manner as we did when selecting the real ELF sample. We calculated the new FIR / radio ratios for the flux-and-luminosity-limited samples, each of which contained ~ 80 of the original ELFs. Such a procedure should be sensitive to subtle biases in the FIR / radio ratio that might possibly be introduced by the flux-density and luminosity limits. We found that the 1000 FIR / radio ratios before applying the FIR flux-density limits had a mean of 2.8400 and a standard error of the mean of 1.4x10-4. After applying the flux-density and luminosity limits, this was unchanged, with a FIR / radio ratio of 2.8395 and a standard error of the mean of 5x10-4. Therefore, our FIR selection method is not expected to bias the sample towards FIR-loud objects.

5.1.8. Future Opportunities

To distinguish between these many competing explanations, some strong observational tests are possible. The heavily obscured compact nuclear starburst model predicts that line emission from ionized hydrogen should be powerful and characteristically concentrated within the central kpc-scale nuclear region. A measurement of the intrinsic line strength measures the total number of ionizing stars and so gives a direct measure of the contribution made by the starburst to the total FIR emission. Such a measurement must carefully avoid the dust obscuration that is invoked to explain the observed shortfall of H emission, and this can be done using near-infrared transitions of hydrogen, or perhaps with a sensitive radio recombination line experiment. The diagnostic power of this approach using the Br line has been demonstrated by Prestwich et al. (1994), and so the technique can be applied to study particular ELFs in question.

Further, the compact nuclear starburst model predicts a distinctive radio spectral shape that is produced by the proposed free-free absorbed nuclear synchrotron source. One would choose a few representative ELFs and make flux-density measurements spanning a broad range of frequency from well below the proposed GHz turnover to well above it. The turnover frequency is expected by Condon et al. to occur at almost a thousand times higher frequency than occurs normally in spirals. This should provide a clear signature, even though the spectral indices on either side of the turnover are not expected to be distinctive.

5.2. Radio Structure

The radio morphology of our ELF sample provides useful constraints on schemes that seek to unify ELFs with AGNs. In particular, Sanders et al. (1988b) proposed, interestingly, that ELFs are the progenitors of optically selected quasars. According to this scenario, the action begins when two normal, gas-rich spirals collide and trigger a nuclear starburst and cause the fuelling of a pre-existing massive central black hole. Both the starburst and the AGN are enshrouded in dust that reprocesses most of the bolometric output into the FIR. This marks the birth of the ELF phase. With time, the nuclear starburst will blow away its cocoon of gas and dust and the full glory of the quasar will be revealed to our optical view. According to this scenario, since 10% of quasars are radio-loud, we might expect 10% of our ELF sample to be progenitors of radio-loud quasars. That we do not see extended powerful radio lobes (Ò?Òš4.4) argues that if some ELFs are progenitors of radio-loud quasars then the radio-loud phase must begin after the ELF phase has finished. Perhaps the dense circumnuclear ISM in ELFs is able to block the formation of large-scale powerful radio lobes and this must be blown away before the jets can emerge.

5.3. Constraining Conditions in the ELF

The nucleus of an ELF is a very different place from the disc of a normal spiral. The gas surface density is an order of magnitude larger, the star formation rates are much larger and produce enormously greater FIR luminosities, and the radio emission is much more intense (e.g., Allen, Roche & Norris 1985; Sanders et al. 1988a; Condon et al. 1991). Quantifying these and other differences is a necessary step towards understanding possible relationships between ELFs and other galaxy types.

Magnetic field strengths profoundly influence radio emission, and help support molecular clouds against gravitational collapse, thus influencing star formation. We can estimate the field strength using equipartition arguments and by using inverse Compton energy loss estimates. The equipartition magnetic field strength is 2x10-8 T, given the observed median Tb of 1000 K at 8.44 GHz for our ELFs, and given the typical source radius of 500 pc, using Equation 13 from Condon (1992). This is much larger than the typical field strengths of ~Òš4x10Ò-10 to 15x10Ò-10 T seen in normal spiral galaxies (e.g., Beck 1992). Remarkably, the following inverse-Compton argument gives a very similar value. To avoid excessive energy losses from the synchrotron electrons by inverse Compton scattering off photons, the energy density in the magnetic field must be larger than that in the photon field (e.g., Melrose 1980). We find that the magnetic field must be > 4x10-8 T if we assume, following Condon et al. (1991), that most of the photons are FIR photons, that LFIR = 1012 L, and that the FIR emission is generated co-spatially with the radio emission (i.e., within the central 500 pc). This is remarkably similar to the equipartition value of 2x10-8 T. These values are similar to the ~Òš10-7 T derived by Condon et al. using the same approach with a similar ELF sample. In such strong fields, the synchrotron lifetime is only 105 yr. This is fairly short compared to the time required for cosmic ray electrons to diffuse appreciable distances away from the central region (105-6 yr per kpc). Thus, we confirm the conclusion of Condon et al. that the radio emission should be nearly co-spatial with the star formation activity, unless local reacceleration is significant.

We speculate that the magnetic field may be large due to adiabatic compression, as the field is frozen into the ISM which is often thought to be driven into the centre by a merger-induced instability. We can estimate the volume of ISM that must have been compressed to cause the observed 100-fold larger magnetic field strength. When adiabatically and homologously compressed (a dubious assumption for ELFs, but the easiest to model), the field strength scales as (radius)-2, assuming flux freezing (following Shklovskii 1960). Thus, the present central 500Òšpc radius slab of dense ISM may previously have been spread over ~Òš5Òškpc. For comparison, the central 5 kpc of the Milky Way contains ~ 9x108 M of ISM (HÒšI + H2) (e.g., Henderson, Jackson & Kerr 1982; Sanders, Solomon & Scoville 1984) which is somewhat less than the typical values of 109 to 1010 M found in the central ~ kpc of ELFs by Sanders et al. (1988a). Thus, there has been more than sufficient ISM compressed into the central region to account for the large magnetic field strengths.

For a starburst model of ELFs to be acceptable, it must be able to reproduce the observed large FIR luminosities for sufficiently long periods, using only the supply of ISM that is known to exist. To test the scenario for our ELF sample, we constructed a simple model starburst with a range of IMF slopes and lower- and upper-mass cut-offs. We estimated the steady-state star-formation rate required to maintain the typical FIR luminosity of 1012 L, and the time required for the burst to consume a given typical amount of ISM. The model assumed that the entire bolometric luminosity of the stars was converted to the FIR, and used an IMF that consisted of a single power-law mass function. The IMF scaled as (mass), where was varied between -3.6 and -1.6, and the low-mass cut-off was varied between 0.3 and 4.8 M, and the high-mass cutoff was varied between 20 and 100 M. We found that, once established, the burst could maintain 1012 L for periods between 1x108 and 5x108 yr given an ISM supply of 1x1010 M, for all combinations of values of these parameters. This agrees well with many previous estimates, reviewed by Telesco 1988, and is much less than a Hubble time.

The total mass of gas tied up in radiating stars at any instant during the burst was generally between 107.5 and 1010ÒšM for most combinations of the model parameters. (Only in the most extreme cases where the steepest IMF combined with the lowest low-mass cut-off did the mass rise much above this, up to 1011.5 M.) Star formation rates were typically 20 to 60ÒšM yr-1, compared to ~ 3 M yr-1 for the Milky Way (Telesco 1988). In general, very reasonable values of gas mass are required to maintain the large FIR luminosities and can do so for useful and realistic lifetimes. Thus, on these grounds, the starburst-ELF model is consistent with observations.

Our starburst model also predicted supernovae at rates between 0.3 and 8 SN yr-1, from which we can constrain the radio energetics. Assuming that each supernova produces the same power as Cas A (1.8x1018 W Hz-1 at 1.49 GHz) and decays exponentially over a period of 104 yr, then the median SN rate (2 yr-1) can only account for ~ 10% of the total radio luminosity that is required to place such ELFs on the observed radio-FIR correlation. This is consistent with a similar calculation by Ulvestad (1982). Thus, something other than the direct emission from SNRs produces the radio emission. Shocks in the ISM can continually reaccelerate electrons, and SNae probably also inject fresh electrons into the ISM, and so they might make up this shortfall. It has been shown previously that there is enough kinetic energy in the supernova ejecta to drive this process.

5.4. Triggering the ELF

Inspecting Sky Survey plate images of our ELF sample reveals that many ELFs are interacting, and show spectacular tidal tails, or appear disturbed. The role of interactions in triggering star formation has been discussed at length, and it seems quite reasonable to expect that interactions lead to the observed activity. In some cases of advanced mergers, such as Arp 220 (Norris, 1988, 1990), we can even see the two nuclei of the progenitor galaxies. However, there are other cases (e.g., Lawrence et al. 1989) where it is not so obvious that interactions are involved. In many cases the ELF is an apparently isolated and rather normal-looking galaxy. Often in these cases where there is no apparent cause for activity, other than the activity itself, the concept of an "advanced merger" is invoked as a trigger. The proposal is plausible, since the obvious signs of tidal disruption are only fleeting transient features of numerical galaxy collisions. At later stages as the nuclei merge closer and slowly spiral inward there may be no obvious outward sign of the interaction. However, the argument for "advanced mergers" is circular (the merger is invoked to provide a trigger for the large FIR output, and the large FIR output is often the only evidence of an advanced merger). To test the "advanced merger" hypothesis one could undertake spectral line imaging of the H I to map out the rotation curves of ELFs. These are expected to be badly disrupted by a merger. The discovery of any single ELF with a normal rotation curve and no outward signs of interaction would present a challenge to the "advanced merger" hypothesis. Such an object may require the existence of another trigger mechanism (e.g., that by Sofue 1991).

6. CONCLUSION

We have observed a sample of 27 ELFs with compact and extended configurations of the VLA, to constrain models of the nuclear activity and its evolution.

We have found clear evidence for AGN activity in some of our ELFs, from optical spectroscopy and from long-baseline interferometric observations of compact high brightness temperature radio cores. However, many show no evidence for AGN activity, but signs of starburst activity are common. Our ability to see light from the NLR and BLR of many ELFs seems inconsistent with the large obscuration that is sometimes invoked to obscure our view of the nuclear starburst. Comparison of optical and NIR transitions of hydrogen may help resolve the conflict.

We found no cases of extended radio-galaxy-like radio lobes produced by ELFs. If ELFs evolve into quasars, then the powerful radio emission must be produced after the galaxies evolve beyond the ELF phase by ejecting the nuclear gas and dust.

Although many ELFs in this sample show optical evidence for interaction, there are a few notable cases that do not, and for these, there may be a considerable problem in explaining what triggered the activity. However, first one must rule out the "advanced merger" hypothesis, and this may best be done using spectral-line observations to map out the rotation curves and to look for abnormalities.

Surprisingly, both our ELFs and those of Condon et al. (1991) fall slightly below the normal radio-FIR correlation. We discussed a range of possible causes, but have yet to find an entirely satisfactory explanation. The compact nuclear starburst model of Condon et al. may explain this offset, although additional effects, such as possible differences between the IMF slopes or the thermal radio fractions of ELFs and normal galaxies, may also contribute to the FIR-excess. We propose observational tests that may rule out some of these models. Heavily obscured nuclear starbursts are expected to be distinctive in near infrared or radio-recombination lines of hydrogen, both of which are able to penetrate through large quantities of obscuring dust. Further, the radio spectrum is predicted to be optically thick up to ~ GHz frequencies according to the compact nuclear starburst model of Condon et al. (1991). Flux-density measurements over a broad frequency range for a few ELFs should yield a strong diagnostic.

ACKNOWLEDGEMENTS

The National Radio Astronomy Observatory is operated by Associated Universities, Inc., under cooperative agreement with the National Science Foundation.

This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, Caltech, under contract with the National Aeronautics and Space Administration, and we are grateful to those who operate this facility.

Table 1: The ELF sample: data

VLA flux density / mJy

Spectral SPTI C-array A/B array

IRAS name Alias Redshift Class / mJy 1.49 GHz 8.44 GHz Tapered 1.49 GHz 8.44 GHz Tapered H EW L8.44 GHz q LFIR

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)

Notes to Table 1:

The first 27 lines give the sample treated in the text. The last 11 lines give galaxies observed but not included in this study. See Ò? 3.1.

Where multiple radio components are present, the flux densities of each component is given.

Ò? Confused

Ò? Close interacting pair

Col. 3: Redshift, mainly from Allen et al. (1991)

Col. 4: Spectral classification from Allen et al. (1991)

Col. 5: 2.3 GHz PTI flux density, or 5- upper limits, in mJy

Col. 6: VLA C-array 1.49 GHz integrated flux density in mJy

Col. 7: VLA C-array 8.44 GHz integrated flux density in mJy

Col. 8: VLA C-array 8.44 GHz integrated flux density in mJy, with array tapered and restored using the 1.49 GHz C-array beam size, for deriving spectral indices.

Col. 9: VLA A/B-array 1.49 GHz integrated flux density in mJy

Col. 10: VLA A/B-array 8.44 GHz integrated flux density in mJy

Col. 11: VLA A/B-array 8.44 GHz integrated flux density in mJy, with array tapered and restored using the 1.49 GHz A/B-array beam size, for deriving spectral indices.

Col. 12: H equivalent width in angstroms, from Allen et al. (1991)

Col. 13: log10 ( 8.44 GHz C-array luminosity / W Hz-1 )

Col. 14: spectral index between 1.49 and 8.44 GHz A/B-array flux densities, cols (9) and (11), where S ~

Col. 15: the FIR / radio ratio at 8.44 GHz using tapered C-array flux densities from col (8), following Helou et al. (1985), q = log10 (( LFIR / W) / 3.75x1012 Hz / L8.44 GHz / W Hz-1 )

Col. 16: log10 ( LFIR / L ) where LFIR is calculated following Helou et al. (1985) and L is 3.83x1026 W

Table 2: PTI data on BGS ELFs

Spectral SPTI

IRAS name Alias class / mJy

(1) (2) (3) (4)

Notes to Table 2:

Col. 3: Spectral class, mostly from Allen et al. (1991) and
VÓ?ron-Cetty & VÓ?ron (1991)

Col. 4: 2.3 GHz PTI flux density or 5- upper limits in mJy

REFERENCES

Allen, D. A., Norris, R. P., Meadows, V. S., & Roche, P. F. 1991, MNRAS, 248, 528

Allen, D. A., Roche, P. F., & Norris, R. P. 1985, MNRAS, 213, 67p

Baars, J. W. M., Genzel, R., Pauliny-Toth, I. I. K., & Witzel, A., 1977, A&A, 61, 99

Barthel, P.D., 1989, ApJ, 336, 606.

Beck, B., 1991, in The Interpretation of Modern Synthesis Observations of Spiral Galaxies, eds N. Duric & P. C. Crane, ASP Conf. Ser., 18, 43

Bregman, J. N. 1990, A&ARev, 2, 125

Buscher, D. 1988, MNRAS, 235, 1203

Chini, R., KrÓÌgel, E., & Kreysa, E. 1992, A&A, 266, 177

Colina, L., & PÓ?rez-Olea, D. 1992, MNRAS, 259, 709

Condon, J. J. 1992, Ann. Rev. A&A, 30, 575

Condon, J. J., Huang, Z. -P., & Yin, Q. F., 1991, ApJ, 378, 65

Condon, J. J., Helou, G., Sanders, D. B., & Soifer, B. T. 1990, ApJS, 73, 359

de Grijp, M. H. K., Miley, G. K., & Lub, J. 1987, A&AS, 70, 95

Devereux, N. A., & Eales, S. A., 1989, ApJ, 340, 708

Devereux, N. A., & Young, J. S., 1990, ApJ, 350, L25

Fernini, I., Burns, J. O., Bridle, A. H., & Perley, R. A. 1993, AJ, 105, 1690

Forbes, D. A., Norris, R. P., Williger, G. M., & Smith, R. C. 1994, AJ, 107, 984

Gioia, I. M., Gregorini L., & Klein, U. 1982, A&A, 116, 164

Hagen, H. -J., Cordis, L., Engels, D., Groote, D., Haug, U., Heber, U., KÓÆhler, Th., Wisotzki, L., & Reimers, D. 1992, A&A, 253, L5

Harwit, M., Houck, J. R., Soifer, B. T., & Palumbo, G. G. C. 1987, ApJ, 315, 28

Haslam, C. G. T., & Osborne, J. L. 1987, Nature, 327, 211

Heckman, T. M., Armus, L., & Miley, G. K. 1990, ApJS, 74, 833

Helou, G., Soifer, B. T., & Rowan-Robinson, M. 1985, ApJ, 298, L7

Henderson, A. P., Jackson, P. D., & Kerr, F. J. 1982, ApJ, 263, 116

Hummel, E. 1981, A&A, 93, 93

IRAS Point Source Catalog, Version 2, 1988, Joint IRAS Science Working Group (Washington, DC: US Government Printing Office)

Isobe, T., Feigelson, E. D., & Nelson, P. I. 1986, ApJ, 306, 490

Kennicutt, R. C., & Kent, S. M. 1983, AJ, 88, 1094

La Valley, M. P., Isobe, T., & Feigelson, E. D. 1992, BAAS, 24, 839

Lawrence, A., Rowan-Robinson, M., Leech, K., Jones, D. H. P., & Wall, J. V. 1989, MNRAS, 240, 329

Lonsdale, C. J., Smith, H. E., & Lonsdale, C. J. 1993, ApJ, 405, L9

Marx, M., KrÓÌgel, E., Klein, U., & Wielebinski, R. 1994, A&A, 281, 718

Melrose, D. B. 1980, Plasma Astrophysics: Nonthermal Processes in Diffuse Magnetized Plasmas Vol. 1, New York: Gordon and Breach, p134

Norris, R. P. 1985, MNRAS, 216, 701

Norris, R. P., 1988, MNRAS,

Norris, R.P., 1990, in IAU Symposium 124: Paired and interacting galaxies, eds. J. W. Sulentic, W. C. Keel, & C. M. Telesco, NASA, Washington, D.C.

Norris, R. P., Allen D. A., & Roche, P. F. 1988a, MNRAS, 234, 773

Norris, R. P., Kesteven M. J., Wellington K. J., & Batty M.J. 1988b, ApJS, 67, 85

Norris, R. P., Allen, D. A., Sramek, R. A., Kesteven, M. J., & Troup, E. R. 1990, ApJ, 359, 291

Paturel, G., FouquÓ?, P., Bottinelli, L., & Gouguenheim, L. 1989, Catalogue of Principal Galaxies, Lyon: Observatoires de Lyon et Paris-Meudon

Peacock, J. A., Miller, L., & Longair, M. S. 1986, MNRAS, 218, 265

Prestwich, A. H., Joseph, R. D., & Wright, G. S. 1994, ApJ, 422, 73

Roy, A. L., Norris, R. P., Kesteven, M. J., Troup, E. R., & Reynolds, J. E. 1994, ApJ, 432, 496

Roy, A. L., Norris, R. P., Kesteven, M. J., Troup, E. R., & Reynolds, J. E. 1995, in preparation

Rowan-Robinson, M. et al. 1991, Nature, 351, 719

Sanders, D. B., Scoville, N. Z., Sargent, A. I., & Soifer, B. T. 1988a, ApJ, 324, L55

Sanders, D. B., Scoville, N. Z., & Soifer, B. T. 1991, ApJ, 370, 158

Sanders, D. B., Soifer, B. T., Elias J. H., Madore, B. F., Matthews, K., Neugebauer, G., & Scoville, N. Z. 1988b, ApJ, 325, 74

Sanders, D. B., Soifer, B. T., Elias, J. H., Neugebauer, G., & Matthews, K 1988c, ApJ, 328, L35

Sanders, D. B., Solomon, P. M., & Scoville, N. Z. 1984, ApJ, 276, 182

Scoville, N. Z., Sargent, A. I., Sanders, D. B., & Soifer, B. T. 1991, ApJ, 366, L5

Seab, C. G. 1987, in Interstellar Processes, eds D. J. Hollenbach & H. A. Thronson, Jr., Reidel, 491

Shklovskii, I. S. 1960, Sov. Astron., 4, 243

Sofue, Y., & Wakamatsu, K. 1991, PASJ, 43, L57

Sopp, H. M., & Alexander, P. 1989, Astrophys. Space Sci., 157, 287

Sopp, H. M., & Alexander, P. 1991a, MNRAS, 251, 112

Sopp, H. M., & Alexander, P. 1991b, MNRAS, 251, 14p

Staveley-Smith, L., Norris, R.P., Chapman, J.M., Allen, D.A., Whiteoak, J.B., & Roy, A.L., 1992, MNRAS, 258, 725

Telesco, C. M. 1988, Ann. Rev. A&A, 26, 343

Ulvestad, J. S. 1982, ApJ, 259, 96

van den Broek, A. C., 1992, A&A, 261, L1

van den Broek, A. C., van Driel, W., de Jong, T., Lub, J., de Grijp, M. H. K., & Goufrooij, P. 1991, A&AS, 91, 61

VÓ?ron-Cetty, M. -P., & VÓ?ron, P. 1991, A Catalogue of Quasars and Active Nuclei, 5th ed. European Southern Observatory, Garching bei MÓÌnchen, Germany.

Wilson, A. S. 1988, A&A, 206, 41

Wink, J. E., Altenhoff, W. J., & Mezger, P. G. 1982, A&A, 108, 227

Wunderlich, E., Klein, U., & Wielebinski, R. 1987, A&AS, 69, 487

FIGURE CAPTIONS

FIG. 1.-A selection of images from the 8.44 GHz A/B-array observations, all reproduced at a common scale. These typify the image quality, and display the two objects that showed structure. All other ELFs were unresolved or were only slightly extended. Peak brightnesses are quoted before any coherence corrections were applied.

FIG. 2.-The 8.44 GHz radio v. FIR luminosity for our ELFs ("" = starburst, "" = Seyfert) and for the BGS ELFs from Condon et al. (1991) (), along with lines that indicate (from bottom to top) the radio-FIR correlation found for the BGS spirals (Condon et al. 1990), for the Wunderlich et al. (1987) spirals, for the Marx et al. (1994) UV-excess-selected spirals, and for the Revised Shapley-Ames spirals (Devereux & Eales 1989). These lines are extrapolated beyond the highest luminosity objects in the control samples. The ELFs differ from the normal correlation, tending to be radio-quiet, or FIR-loud.

FIG. 3.-The fraction of galaxies from Allen et al. (1991) classified spectrally as AGN v. FIR luminosity. The AGN fraction increases dramatically at the highest luminosities. The number above each column gives the total number of objects in each luminosity bin.

FIG. 4.-A/B-array core spectral index between 1.49 and 8.44 GHz v. FIR/1.49 GHz C-array radio ratio, q, for the 23 ELFs that had all the required flux densities measured. The nuclear radio spectrum tends to become flatter as ELFs become more radio-quiet. (When calculating the spectral index, the array was tapered at 8.44 GHz to match the resolution of the 1.49 GHz A/B-array observations.)