COLA. III. RADIO DETECTION OF ACTIVE GALACTIC NUCLEUS IN COMPACT MODERATE LUMINOSITY INFRARED GALAXIES

Snapshot radio continuum observations for the COLA-N sources were obtained using the VLA at 4.8 GHz in two observing epochs each of 24 hr in length (see Table 2). Sources with 0° < δ < 30° were observed on 1998 June 23 in the VLA BnA configuration, while sources with δ>30° were observed on 2003 July 19 in the A configuration. In addition, to ensure that all compact sources were observed with the A-array, 12 sources found to be compact at 4.8 GHz from the first epoch were re-observed in the second epoch. The data were reduced using AIPS in a standard manner. Radio maps of size 50'' × 50'' centered at the IRAS positions were produced for all the sources using natural weighting. Typical FWHM beam-sizes at 4.8 GHz were ∼04 and 09 × 04 for A and BnA configurations, respectively. The typical rms noise levels were 75 μJy beam⁻¹

VLBI 4.8 GHz snapshot observations at data rate 1 Gbit s⁻¹ per station were conducted for the COLA North sample using the most sensitive telescopes available in the European VLBI network (EVN). The first epoch on 2005 February 27/28 used Eb (Effelsberg), Wb (Westerbork), and Ar (Arecibo) for all sample sources within the declination range 10° < δ < 40° (defined by the Ar zenith-distance restrictions and minimum elevation at the European stations) with ∼15 minutes of on-source time per target giving σ = 25 μJy on the Ar–Eb baseline. The second epoch on 2005 May 6/7 used Eb, Wb, and Jb1 (Jodrell Bank, Lovell telescope) for all the sources with 0° < δ < 10° and δ>40° with ∼15 minutes of on-source time per target giving noise σ= 56 μJy on the Eb–Jb1 baseline. Gaps in the schedule were filled by re-observing 21 sources from the first epoch. In both epochs, strong nearby phase reference and amplitude calibrators (>500 mJy) were observed for 10 minutes every ∼1 hr. The Eb–Wb baseline (266 km) which is common to both epochs has a fringe spacing of 46.5 mas at 4.8 GHz, which is similar to the 47.4 mas fringe-spacing of the ATCA–Tidbinbilla baseline (566 km) at 2.3 GHz used by Corbett et al. (2002) to observe the southern sample. Because of scheduling restrictions and bad data we only obtained useful VLBI observations for 90 sources.

AIPS was used for the initial data reduction and calibration stages with instrumental phase and delay being determined toward strong calibrators. Monitored system temperatures and gains were used to calibrate amplitudes. An adjustment of ∼30% was required on Jb1 to obtain consistent calibration on sources known to be unresolved from the VLBA calibrator database. Our final estimate is that quoted baseline flux densities are accurate to ∼15%.

Fringe fitting was performed in AIPS on all the calibrators and then these solutions were applied to the interleaved target source scans. After this we exported the data to our own software within which Delay–Rate maps were made for each baseline/target source. The size of these delay-rate maps were set such that a compact source would be detectable up to 20'' from the correlation position. Finally, the Delay–Rate maps were transformed into R.A.–decl. offset maps relative to the phase center. These maps show only features which are smaller than the baseline fringe spacing (see Parra 2007). We adopted a detection threshold of 6 times the measured rms noise.

In addition to our own radio data our analysis made use of source 1.4 GHz total flux densities taken from the NVSS catalog (Condon et al. 1998). Infrared total flux densities were taken from the IRAS Point Source Catalog (Beichman et al. 1988). In order to calculate source infrared to radio (1.4 GHz) ratios (q_1.4) as defined by Helou et al. (1985) we calculated for each source an equivalent FIR flux density from S_FIR = 1.26 × 10⁻¹⁴(2.58f₆₀ + f₁₀₀)/3.75 × 10¹² W m⁻² Hz⁻¹, where f₆₀ and f₁₀₀ are the cataloged 60 and 100 μm IRAS flux densities. For comparison with COLA-S and other data sets for each source a FIR spectral luminosity (see Table 1, Column 9) was calculated assuming isotropic emission. Finally, bolometric IR (8μm–1000μm) luminosities in solar units were calculated (see Table 1, Column 7) from the IRAS flux densities using the relation given by Sanders & Mirabel (1996).

Optical spectroscopy observations were taken using the 60 inch CfA telescope at Mt Hopkins, Arizona between 2001 February and October using the FAST spectrograph. A slit width of 3'' was used, corresponding to 0.9 kpc at the median sample source distance. A grating of 300 lines/mm was used to disperse light between 4000 and 7000 Å and gave a spectral resolution of 6 Å. A single exposure of between 900 and 1200 s was made for each target. Spectrophotometric and smooth standards (often the same stars) were observed during each night and used to flux-calibrate and correct for atmospheric absorption features in the spectra. We were able to measure useful line ratios for 61% of the sample.

Data reduction was carried out in the standard manner using IRAF. The frames were de-biased and flat fielded. Galaxy spectra were extracted using an aperture centered on the position of peak flux across the slit. If more than one peak was seen in the spatial direction a spectrum was extracted at the position of each peak. Sky-subtraction was then carried out and wavelength calibration done using an Ne–Ar arc lamp. The observations were then corrected for atmospheric extinction and flux calibrated using the spectrum of a spectrophotometric standard. Finally, atmospheric absorption features in the target spectra were removed using the smooth spectrum standard observed at a similar air mass to the target.

The line fluxes were measured by fitting (multiple) Gaussians to the emission lines using NGAUSSFIT in IRAF. The continuum flux was subtracted using a linear fit to small portions of the continuum either side of the emission line. The Hα (λ 6562 Å) and [N ii] doublet (λλ 6583, 6548 Å) were often overlapping, in these cases line profiles were de-blended by simultaneously fitting a separate Gaussian for each line. When Hα was present in absorption as well as emission, both the emission and absorption features were fitted simultaneously with Gaussians.

In a FIR-selected sample of galaxies such as COLA, it is expected that in most sources star formation will be the dominant mechanism powering the FIR emission. For such sources radio and FIR luminosities and flux densities are expected to be tightly correlated (see Condon 1992). The total 1.4 GHz and FIR luminosities for our sample are plotted in Figure 1 (left). Except for one outlier (3C84) which is unusually radio luminous, sources tightly follow the expected correlation. We calculated for each source the logarithmic ratio of flux densities q_1.4 = log(S_FIR/S_1.4) as defined by Helou et al. (1985). Excluding 3C84, the mean value of q_1.4 is 2.34 ±  0.01 (and the median 2.31 ±  0.01) which is in good agreement with the mean q_1.4 = 2.34 ±  0.01 measured by Yun et al. (2001) from a sample of 1809 galaxies (solid line in Figure 1). In contrast the median value for COLA-S (Corbett et al. 2002) is q_1.4 = 2.43 ±  0.03 which is significantly (4σ) larger than the COLA-N median value.

To try to resolve the above discrepancy we compared NVSS and COLA fluxes for the 56 COLA South sources also contained in the NVSS catalog (i.e., those with δ ⩾ −40°). We found NVSS flux densities for these sources which were 12% larger than those measured by Corbett et al. (2002), resulting in q_1.4 = 2.36 ±  0.05 for ATCA fluxes and 2.31 ±  0.04 for NVSS fluxes. It has been noted by Prandoni et al. (2000), Norris et al. (2006), and Kellermann et al. (2008) that for sources with flux densities ≲15 mJy NVSS flux densities tend to be ∼10% larger than ATCA measurements. Although COLA total flux densities are about a factor of two above this flux limit the effect is neither understood nor well defined and so this scale difference may contribute to the apparent difference of median q_1.4. We conclude that the difference in q_1.4 between COLA-N and COLA-S is likely caused by (1) a ∼10% calibration difference for weak sources between NVSS and ATCA flux densities, combined with (2) a small number of sources at δ < −40° which have anomalously large values of q_1.4, perhaps simply because of small number statistics.

It remains unclear whether the likely ∼10% calibration difference between Northern and Southern hemispheres noted above for weak sources is a property of NVSS specifically or all VLA observations. Furthermore, the 4.8 GHz measurements of galaxies made with the Bonn 100m single dish give q_4.8 = 2.71 ±  0.03 (Wunderlich et al. 1987), assuming a spectral index of 0.75 this then gives an estimated q_1.4 = 2.31 ±  0.03 close to the VLA value, supporting the idea that Northern hemisphere q_1.4 values are consistent with each other. Given the importance of sorting out the origin of radio flux scale differences between hemispheres it seems that dedicated simultaneous ATCA/VLA observations of sources with different flux densities should be made.

The results of our VLA observations are summarized in Table 3. We successfully detected and mapped 95 out of the 110 observed sources. The 15 non-detections are probably large galaxies with a surface brightness too low to be detected with the VLA configuration used.

The images for the VLA detections are shown in Figures 2 and 3 with the former showing sources which are also VLBI detections and the latter showing those that are VLBI non-detections. The images show a wide variety of morphologies. Following Neff & Hutchings (1992) we have classified the sources into four different morphological classes: Core (3), Core and Extensions (CE), Extended with no core (E), and Undetected (U), each class containing 28, 62, 5, and 15 sources, respectively. There are two sources with clear ring structures, namely IRAS05091+0508 (NGC 1819) and 23488+1949 (NGC 7771); there are an additional four showing elongated linear structures typical of edge-on disks.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View extended figure (7 panels)

Total recovered fluxes at 4.8 GHz (S_4.8total) were obtained by integrating the flux over a region enclosed by the lowest reliable contour of radio emission. The median q_4.8 = log(L_FIR/L_4.8tot) obtained for the sample is 3.03 ±  0.05 which is significantly higher than the 2.71 ±  0.03 measured by Wunderlich et al. (1987) using single dish observations of a sample of 99 normal spirals. We conclude that we are on average resolving out ∼50% of the 4.8 GHz flux density. We note however that for those sources with compact central features the fraction of missing flux density is ≲20%.

We detect, on at least one VLBI baseline, 20 out of 90 sources observed. The measured flux densities on each baseline are given in Table 4. On the shortest Eb–Wb baseline (see Figure 4) the signal-to-noise ratio for most detections is generally significantly larger than the 6σ detection threshold (see Figure 4) while non-detections are bounded to <4σ. It is notable that we obtain VLBI detections for five of the six sources with IR luminosities above 10^11.4 L_☉. A histogram showing the VLBI detections and non-detections versus IR luminosity is presented in Figure 5 (top) which shows that the probability of detecting a VLBI core increases with IR luminosity.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4 shows that there are nine sources with simultaneous measurements on the Eb–Wb and Eb–Jb1 baselines. Of these, seven have similar fluxes (within the noise) on these two baselines, the latter of which is ∼2.6 times longer than the first, showing that the VLBI component is unresolved.

Of the 14 sources with simultaneous Eb–Wb and Eb–Ar observations, 9 are detected on the Eb–Ar baseline, but with significantly reduced flux density, indicating that the VLBI sources are resolved on the longest baseline. We estimate an angular size (see Table 4) by approximating the flux ratio between both baselines to that expected from a circular Gaussian (see Pearson 1999). The high sensitivity and small fringe spacing of the Eb–Ar baseline resulted in a very small uncertainty on the phase center offsets of the detections. These VLBI position offsets are shown as crosses on the corresponding VLA map in Figure 2.

Figure 1 (left) shows that the total 1.4 GHz VLA flux densities of our VLBI detected sources follow the FIR to radio correlation. There are 15/20 detections within ±0.3 dex of the empirical line defined by Yun et al. (2001) (i.e., within the gray region shown). Of the remainder two are just slightly above and two slightly below the ± 0.3 dex line. Only one source, 3C84, is significantly above the correlation. At 4.8 GHz a similar trend is observed with 14/20 detections located within ±0.3 dex of the empirical line defined by Wunderlich et al. (1987). Consistent with the above results we find the median q_1.4 for the VLBI detections and non-detections to be indistinguishable (i.e., 2.30 ±  0.08 and 2.31 ±  0.02 respectively). Likewise, if we omit 3C84 the mean of the VLBI detection is 2.33 ±  0.08 and of the non-detections is 2.32 ±  0.02. We can conclude that the data show that excepting 3C84 the total radio flux density is dominated by star formation regardless of whether or not a VLBI core is present.

The above results appear to differ from those in COLA-S where a difference in median q_1.4 between VLBI detections and non-detections was claimed. From Table 3 of Corbett et al. (2002) the median q_1.4 value and its estimated one sigma error for the nine definite COLA-S VLBI detections was 2.07 ±  0.17 while for the VLBI non-detections it was 2.44 ±  0.03. The errors on theses estimates are calculated from the internal dispersion about the median q_1.4 for the two sub-samples as given in Corbett et al. (2002). We ascribe this apparent two sigma difference between the two sub-samples in the South as being due to the lower VLBI sensitivity in the South combined with the overall form of the q_1.4 distribution. This distribution has a tail of radio core-dominated, low q_1.4 sources (such as 3C84 in the North and NGC5793 in the South) which are easily VLBI detected plus weaker VLBI cores in galaxies whose radio emission is star formation dominated. The lower sensitivity VLBI observations in the South contain a larger fraction of the former class of source which reduces the estimated q_1.4 of VLBI detection compared to non-detections. Specifically, in the South four out of nine VLBI detected sources have q_1.4 ≲ 2 while the fraction is only 3 out of 20 in the North. Because low q_1.4 sources comprise almost half of the VLBI detected sample in the South they can bias the median q_1.4 while in contrast in the North such outlier sources have negligible effect.

Visual inspection of the VLA radio maps indicates a bias for VLBI detections to be in sources with compact structure and high peak brightness (see Figure 2). This is confirmed in terms of VLBI detection rate versus morphology class (see Section 4.2). Among the VLBI observed sources there are 13/27 detections of class C, 6/48 of class CE, and 1/5 of class E. We do not detect any class U source. This result was confirmed by plotting VLA visibility amplitude versus uv distance. The majority of VLBI detections were indeed found in sources with compact, but not unresolved, arcsecond-scale structures. The size of these structures was estimated by subtracting an unresolved source equivalent to the VLBI detection from the brightest component in each VLA map and then fitting a Gaussian to the difference.

In two cases (00005+2140 and 06239+7428), the VLBI core subtraction removed all the flux density implying that the brightest VLA component was caused completely by the VLBI source itself. We consider these two sources together with 03164+4119 (3C84) to be VLBI core-dominated sources and omit them from the subsequent analysis. For the rest of the sources the task JMFIT was used to fit a Gaussian within a tight box containing the remaining feature. For the vast majority of sources in classes C and CE with a clear compact component there was no ambiguity in the Gaussian fitting, i.e.,, the results of the fits were the same regardless of the size of the fitting box. For sources in class E, the size estimates were found to vary by up to a factor of ∼2 depending on the chosen box. However, in all such cases the fitted sizes were large (≳2''). The resulting component flux densities and major and minor axes θ_M and θ_m are listed in Columns 6 and 7 of Table 3.

Peak brightness temperatures T_b at 4.8 GHz for the compact VLA structures were calculated using the relation T_b ≃ 35 × (S_4.8 peak)(θ_M × θ_m)⁻¹ K° (Column 8 in Table 3). In a few sources the fitted deconvolved minor axis was zero; in these cases its size was set to 50 mas which was equal to the smallest size found in other sources.

Uniquely, among sources with compact structure we did not collect any VLA A-array data for the well known ULIRG source 15327+2340 (Arp220). In order to treat this source in a similar way to the rest we give in Table 3 radio component properties from the literature for the brighter western nucleus (sizes from Condon et al. 1991 with flux densities scaled to 5 GHz using the spectral index of Rovilos et al. 2003).

As can be seen from Figure 5 (bottom panel) VLBI detections seem to be strongly correlated with those sources whose VLA fitted components (after removing any VLBI contribution) have the highest T_b. VLBI detectability is less strongly correlated with large L_IR (top panel). Two-sided Kolmogorov–Smirnov tests comparing the distribution of VLBI detected and non-detected sources in L_IR and T_b in each case rejects the hypothesis that VLBI/non-VLBI sources are drawn from the same population at probability levels of P = 0.82 and P = 0.98, respectively. This result demonstrates that the stronger correlation for VLBI detectability is with T_b rather than with L_IR.

To classify sources based on their line ratios from our optical spectroscopy observations (see Section 3.4) we used the method of Kewley et al. (2006). This scheme first classifies sources into H ii galaxies, AGN/star formation composites and pure AGN based on a mixture of theoretical (maximal starburst, Kewley et al. 2001) and empirical line-ratio boundaries (Kauffmann et al. 2003) delineating pure star-forming galaxies. These boundaries are shown respectively as solid and dashed curves delimiting the bottom left corners of the diagnostic line ratio diagrams in Figure 6. H ii galaxies (class H) are to the left and below all the curves including the dashed line in Figure 6(a). Composite sources (class C) lie between the solid and dashed lines in the [N ii]/Hα diagram and AGN are above and to the right of the curves in all diagrams. The AGN population is further divided between Seyfert (class S) or LINER (class L) using the [SII]/Hα and [O i]/Hα diagnostic diagrams and the empirical results of Kewley et al. (2006) who found two well defined populations tracing different position angles on these diagrams. For a significant number of our sources either the [SII] or [O i] lines were not detected. In order to increase the number of detections we therefore used the same 2-of-3 criterion for classification as Yuan et al. (2010) whereby if there was no data for a given line ratio or if one diagram gave a classification inconsistent with the other two we chose the classification given by the majority of the diagrams.

Zoom In Zoom Out Reset image size

Our available optical data allowed for the classification of 66 sources as summarized in Column 10 of Table 3. The fraction of different classifications agrees well with that found among sources of similar luminosity by Yuan et al. (2010). For the subset of sources observed by VLBI the compact radio core detection rates for each optical category are; Seyferts 4/8 (50%), LINERS 4/6 (66 %), Composite 3/20 (15%), and H ii 2/15 (14%). These VLBI detection fractions are broadly consistent with the idea that VLBI cores trace AGN activity since the detection fraction is largest among AGN optical types (see Section 5.4).

The primary objective of our VLBI observations was to detect compact cores indicating AGN activity. Here we discuss whether our 20 VLBI detections can be taken as unambiguous signs of AGN activity or whether they might be confused by the high spatial frequency tail of arcsecond scale diffuse emission, radio SNe, or supernova remnants (SNRs).

5.1.1. VLBI Brightness Temperature and Diffuse Starburst Emission

5.1.2. Compact Sources in Arp 220

5.1.3. Supernova Remnant Origin

5.1.4. Radio Supernova Origin

For most VLBI detected COLA sources (17/20), a large fraction of the total radio flux density arises from several 100 pc scale structures seen in the VLA maps. As noted in Section 4.1 the total flux densities of nearly all COLA sources, including both VLBI detections and non-detections (see Section 4.4 and Figure 1), closely follows the FIR to radio correlation at both 1.4 and 4.8 GHz. This result indicates that most of the radio emission recovered in the VLA maps is powered by star formation activity. It is known from observations of the starbursts in M82 and Arp 220 that the associated star formation induced compact emission from SNR and SNe is just a few percent (Lonsdale et al. 2006) of the total diffuse component radio emission; hence there is no contradiction between star formation being the main power source for the total radio emission and our conclusion that it does not provide enough correlated flux density to explain the VLBI detections.

Figure 7 shows a IR color–color diagram for the 90 sources with VLBI observations in the COLA North sample. Any position on this diagram can be reached by a linear combination of starburst temperature, AGN and reddening (Kewley et al. 2000). It is therefore impossible to unambiguously estimate the proportions in which these ingredients are combined. However, most of the sources in the sample appear to follow the line traced by blackbodies of progressively increasing temperatures (i.e., the starburst line) with a scatter apparently increasing with temperature suggesting a larger AGN contribution to the IR energy budget in warm starbursts. The overall scatter of the COLA sample is however much smaller than that observed by Rush et al. (1993) in their sample of optically selected Seyferts. We therefore argue that for most sources in our sample that their IR emission is mostly powered by starburst activity with any AGN contribution being energetically unimportant.

All 10 VLA non-detections (see Section 4.2) with VLBI data are clustered near the cold end of the starburst line (crosses in Figure 7). As discussed in Section 4.2 these are likely to be large sources whose relatively large FIR luminosity is not due to concentrated nuclear starburst activity but is instead caused by the aggregation of normal star forming regions spread over their large galactic disks.

As discussed in Section 4.6, we find that the fraction of VLBI detections in AGN-like optical types is substantially higher than in H ii type galaxies, supporting the interpretation that VLBI detections indicate AGN activity. We find however no one-to-one correspondence between the optical classification and VLBI detection. The fact that half of the observed Seyfert sources (4/8) were not detected by VLBI supports the result of Corbett et al. (2003) who found observationally two classes of Seyferts with and without detectable VLBI radio cores. This could simply be an artifact of our limited VLBI sensitivity or it could mean that there are physically two classes of Seyfert with different degrees of radio-loudness. The relatively high VLBI detection rate of LINER sources (4/6) is consistent with models in which this class is usually AGN powered (Kewley et al. 2006) although it is clear that in some sources LINER-like spectra can arise in shocks, i.e., for the case of Arp 220 (Arribas et al. 2001). The relatively low incidence of VLBI detections in H ii galaxies (2/15) suggests that cases of completely optically obscured AGN are rare. Combining the VLBI and optical results we find the overall incidence of AGN activity to be high. If we add the four non-VLBI detected Seyferts to the 19 VLBI detections (excluding Arp 220) we get a minimum fraction of sources containing AGN of 22/90 (25%).

An interesting correlation found in the present work is that between the presence of a VLBI detection and of high brightness arcsecond-scale radio emission (see Section 4.5 and Figure 5). In Section 5.1 we argued that VLBI detections are due to AGN activity rather than SNe or SNRs. In Section 5.2, we further argued that the extended radio emission seen with the VLA is due to star formation. Taken together these results imply a strong connection between AGN activity and the presence of a compact starburst.

Zoom In Zoom Out Reset image size

A significant body of evidence in favor of a strong starburst/AGN connection has been accumulating in recent years. In nearby AGNs evidence is seen for for nuclear post-starburst stellar populations of age ∼10⁶–10⁸ year (Davies et al. 2007) suggesting a possible evolutionary connection with nuclear starbursts preceding AGN activity (see also Heckman 2008). There are however also many individual examples of concurrent starburst/AGN activity (e.g., Fathi et al. 2006). The present paper has found evidence linking AGN activity with concurrent compact starbursts in FIR selected sources and it is interesting to consider possible physical interpretations of this result.

The Eddington-limited starbursts discussed by Thompson et al. (2005), which can also feed central AGN, are predicted to have central peak bolometric brightness of 10¹³ L_☉ kpc⁻² consistent with the luminous compact starbursts from Condon et al. (1991). Figure 8 plots the estimated FIR surface brightness versus fitted size for the arcsecond-scale radio structures for our COLA sources, matching Figure 4 of Thompson et al. (2005). We find that our VLBI sources are mostly detected in sources with bolometric brightness approaching or around 10¹³ L_☉ kpc⁻² which coincides with values predicted for Eddington-limited starbursts.

Zoom In Zoom Out Reset image size

Thompson et al. (2005) also argued for two separate modes by which such compact starbursts can feed a central black hole, supplying either 0.1 M_☉ year⁻¹ or 4M_☉year⁻¹ depending on whether the gas supply rate at several hundred parsec radius is smaller or larger than 220M_☉yr⁻¹. The rest of this mass is converted into stars. Since the star formation rates of COLA sources are ≪220 M_☉yr⁻¹ the lower black hole supply rate applies. Assuming an efficiency of 10% this supply rate translates into an AGN bolometric luminosity of 1.5 × 10¹¹L_☉ which is comparable to the IR luminosities of the COLA galaxies with VLBI detections. Such a high AGN luminosity, which would mostly emerge in the IR, is inconsistent with the fact that COLA sources lie on the FIR-radio correlation and have IR colors typically found in starburst-powered sources (see 5.3). We can speculate that a constant fraction of gas supplied to the black hole might be lost to winds or jets formed close to the black hole event horizon, thus reducing the AGN bolometric luminosity. The detected core radio emission would be generated at the base of these jets (Falcke & Biermann 1995). The starburst regulation of the gas supply would ensure that these jets had similar mass outflow rates and mechanical luminosities therefore explaining the relatively narrow range of radio luminosity of our VLBI detections.

The high incidence we find of AGN activity in compact high brightness starbursts (i.e., VLBI detections in 13/18 sources with specific star formation rates >10¹² L_☉ kpc⁻², see Figure 8) suggests only a short interval between the triggering of such starbursts and the onset of AGN activity. In contrast Davies et al. (2007) argue (based on AGN stellar population studies) that there is a 50–100 Myr delay between the peak of nuclear starburst activity and the onset of AGN activity. These studies were however done on AGN selected sources. AGNs plausibly last much longer after the initial feeding event than does starburst activity. Possibly, this is because once the central ∼1 pc scale accretion disk is fueled with gas via a compact starburst long viscous accretion times (up to 10⁹ yr. King & Pringle 2007) allow AGN activity to continue long after starburst activity has faded. Given this timescale difference examples of currently active nuclear starbursts in AGN selected samples might well be expected to be rare. Additionally, during the initial composite AGN/starburst phase optical obscuration may be very large making such objects hard to be optically identified as AGN. Since COLA is FIR selected and we have used radio observations as the primary means to detect AGN both of the above biases are avoided allowing the detection of the initial phases of composite AGN/starburst activity.

Both our own observations and those of Condon et al. (1991) show a strong correlation between high FIR luminosity and compact high brightness starbursts. Within the IRAS-BGS sample Condon et al. (1991) found that all starbursts with L_FIR>10^11.8 L_☉ were in fact of this high brightness Eddington-limited type. If the correlation between AGN powered VLBI sources and high brightness shown in Figure 8 also holds for this more luminous sample then one would expect all of these to have AGN activity detectable via VLBI. It can be argued that Arp 220 is a clear counter-example since it has high radio brightness and although detected by VLBI this detection is due to SNe and SNRs rather than an AGN core or lobes. However, as noted by Parra et al. (2007) within the western nucleus of Arp 220 there exist several candidate sources for a radio AGN. Recent millimeter continuum and ¹²CO(2–1) spectral line interferometry results by Downes & Eckart (2007) and Sakamoto et al. (2008) strongly bolster the case that a powerful AGN does in fact exist in the western nucleus of Arp 220.

More generally, within snapshot 1.4 GHz VLBI observations conducted by Lonsdale et al. (1993), about 50% of luminous IRAS-BGS sources were detected on global baselines. These detections had however ambiguous interpretations being explainable by both AGN and radio SNe/SNR models. In general, at other wavebands attempts to definitely detect or reject AGN contributions in these very luminous starbursts have also often proved inconclusive. If as predicted by the Thompson et al. (2005) model for Eddington-limited starbursts the gas supply to the central black hole is relatively constant for compact starburst sources we may obtain the same AGN luminosity regardless of L_IR. It follows that in more intense higher IR luminosity starbursts it may be harder to detect the AGN component in the midst of all the starburst activity. Paradoxically it may be that within the moderate luminosity COLA sample a strong correlation between intense starburst activity and radio AGNs is easier to detect than in more luminous samples. This is firstly because COLA straddles a luminosity range were there are large numbers of both low brightness and Eddington-limited starbursts. Second, L_IR is not high enough (except for Arp 220) that there is significant confusion in the radio between AGN and starburst induced structure.

It is clear that both the VLBI snapshot observations of COLA-N presented in this paper and the observations of IRAS-BGS sources carried out by Lonsdale et al. (1993) suffer from lack of uv coverage. New long track VLBI observation which produce good quality images would help distinguish star formation origins of VLBI flux density (by for instance seeing multiple compact VLBI components indicating star formation, e.g., Parra et al. 2007) from AGN origins (core-jet morphology sources). Observations of half of the VLBI detected COLA-N sources have already been made using the EVN at 6 cm wavelength. These data are presently being reduced.