Authors: Kunal Mooley (Caltech), Dale A. Frail (NRAO)

Our knowledge of the variable GHz sky is especially lacking at sub-mJy flux density levels, where star forming galaxies and radio-quiet AGNs emerge as dominant source populations. Majority of the objects seen by the VAST-Deep surveys will belong to these, mostly unexplored, populations. Is the variability in the sub-mJy sky dominated by AGNs or are we probing star-formation? How many variables and transients can the future wide-field surveys expect to find? What is the rate of transient phenomena? Given this background of false positives, what is the optimum frequency to search for electromagnetic counterparts to gravitational waves? Keeping in mind the large data rates from all VAST surveys and the necessity of automated transient search, understanding the completeness and reliability of our source catalogs is crucial. How do source-finding algorithms differ from one another? What are the optimum input parameters to use for transient search? These are some of the issues that we have addressed in Mooley et al. (2013, ApJ, 768, 165; arXiv:1303.6282).

In Mooley et al. (2013), we present a search for variables and transients using archival 1.4 GHz VLA data taken as part of a deep radio continuum survey toward the Extended Chandra Deep Field-South (E-CDFS), a 34' × 34' region. One of the advantages of working with archival data taken toward these deep fields is there is plenty of ancillary data at other wavelengths. For any interesting variable or transient we have deep X-ray, optical, and infrared observations, allowing us in many cases to identify the properties of the quiescient host.

The 245 hours of data were taken toward E-CDFS, consist of 49 epochs, each achieving an rms of about 30 μJy, and together sample timescales of one day to three months. We find that only a fraction (1%) of unresolved radio sources above 40 μJy are variable at the 4σ level. There is no evidence that the fractional variability changes along with the known transition of radio source populations below one mJy. Optical identifications (Figure 1) of our variable sources show that the variable radio emission is associated with the central regions of an active galactic nucleus or a star-forming galaxy. The variability is likely due to the central engine, but gamma-ray bursts (GRBs), super-luminous Ib/c or IIn supernovae, while rare, could be responsible. Such objects could be present in the compact starburst regions observed in the central kpc regions of galaxies, irrespective of whether they are AGN hosts (e.g. Brunthaler et al. 2009, A&A, 499, 17).

Figure 1: 2.2" × 2.5" HST (F606W) image cutouts of variable sources found in E-CDFS. The red error ellipses denote the radio source positions shift-corrected to the HST source positions using 1σ error. The redshift (z), peak luminosity (L) and absolute R-band magnitude (M) are also given. SF denotes a star-forming galaxy and AGN denotes an active galactic nucleus as evidenced from the radio and mid-to-far infrared data available for these objects.

For any transient search, it is important to reliably distinguish noise from real transients, and to avoid the rejection of transients as noise. Thus, characterizing the reliability and completeness of source catalogs is crucial. Due to the availability of a deep continuum image and a painstakingly constructed catalog of E-CDFS sources, we were able to test the efficacy of various source-finding algorithms on radio interferometric data. We find significant differences between the algorithms in terms of components fitted for extended sources, sidelobe rejection, and point sources detected in regions where the rms is appreciably larger than the mean rms. Some of these differences are illustrated in Figure 2. For applications that need both completeness as well as reliability, sfind and Aegean are good. IMSAD also gives a good completeness and reliability for detection thresholds of 6--10σ. For transient searches, reliability takes preference over completeness, since false positives are likely to consume follow-up resources. We find that the best reliability is provided by SAD. None of the frequently used algorithms in radio astronomy are able to characterize multi-component sources perfectly well. While IMSAD tends to fit a single component to any extended source, sfind, SAD and Aegean often fit too many, most of which seem unphysical. SExtractor may be better in this sense, but it has other drawbacks, especially in terms of reported source positions and completeness. IMSAD and SExtractor catalogs are also plagued with a significant number of false sources in cases where the rms noise varies significantly over the image. If there is sufficient interest for testing search algorithms on real interferometric data, please let us know and we will try to make the E-CDFS data available for use.

Our findings suggest that the radio sky at 1.4 GHz is relatively quiet. Comparison of our study with literature suggests that radio variability is a strong function of frequency. As a guideline for VAST surveys, we can estimate that above 1 mJy there would be one strong variable (having fractional variability larger than 50%) per square degree of the sky, and above 100 $\mu$Jy there would be about 9 strong variables. For multi-wavelength transient searches, such as the electromagnetic counterparts to gravitational waves, 1.4 GHz may be optimal for reducing the high background of false positives.